diff --git a/analysis/current/analysis-report-hybrid.json b/analysis/current/analysis-report-hybrid.json
index 240c0c4bb4b8dc99b56cf3274590bf087cc9752d..886eae36670f9bc7a423b6b301ff1618a2deb775 100644
--- a/analysis/current/analysis-report-hybrid.json
+++ b/analysis/current/analysis-report-hybrid.json
@@ -1,8 +1,8 @@
 {
   "schema_version": "1.0",
   "repo": "huggingface/transformers",
-  "snapshot_id": "20260510T000023Z",
-  "generated_at": "2026-05-10T00:05:56Z",
+  "snapshot_id": "20260510T060019Z",
+  "generated_at": "2026-05-10T06:03:17Z",
   "evidence_quality": "full",
   "llm_enrichment": true,
   "meta_bugs": [
@@ -1148,204 +1148,6 @@
         }
       ]
     },
-    {
-      "cluster_id": "cluster-43656-4",
-      "summary": "Cluster of 1 issues and 3 PRs centered on issue #43824.",
-      "status": "open",
-      "confidence": 0.8,
-      "canonical_issue_number": 43824,
-      "canonical_pr_number": 43656,
-      "issue_numbers": [
-        43824
-      ],
-      "pr_numbers": [
-        43656,
-        43836,
-        43842
-      ],
-      "evidence_types": [
-        "closing_reference",
-        "shared_issue_target",
-        "soft_similarity"
-      ],
-      "pr_comparisons": [
-        {
-          "left_pr_number": 43656,
-          "right_pr_number": 43836,
-          "code_similarity": 0.36,
-          "size_similarity": 0.176,
-          "file_overlap": 1.0,
-          "area_overlap": 0.051,
-          "patch_similarity": 0.048,
-          "shared_filenames": [
-            "src/transformers/cli/serve.py"
-          ],
-          "shared_file_areas": [
-            {
-              "filename": "src/transformers/cli/serve.py",
-              "left_ranges": [
-                [
-                  11,
-                  18
-                ],
-                [
-                  30,
-                  36
-                ],
-                [
-                  315,
-                  323
-                ],
-                [
-                  665,
-                  671
-                ],
-                [
-                  931,
-                  937
-                ],
-                [
-                  1843,
-                  1849
-                ],
-                [
-                  1868,
-                  1874
-                ]
-              ],
-              "right_ranges": [
-                [
-                  11,
-                  18
-                ],
-                [
-                  359,
-                  429
-                ],
-                [
-                  584,
-                  590
-                ],
-                [
-                  1892,
-                  1910
-                ],
-                [
-                  1917,
-                  1923
-                ]
-              ]
-            }
-          ]
-        },
-        {
-          "left_pr_number": 43656,
-          "right_pr_number": 43842,
-          "code_similarity": 0.405,
-          "size_similarity": 0.5,
-          "file_overlap": 1.0,
-          "area_overlap": 0.0,
-          "patch_similarity": 0.036,
-          "shared_filenames": [
-            "src/transformers/cli/serve.py"
-          ],
-          "shared_file_areas": [
-            {
-              "filename": "src/transformers/cli/serve.py",
-              "left_ranges": [
-                [
-                  11,
-                  18
-                ],
-                [
-                  30,
-                  36
-                ],
-                [
-                  315,
-                  323
-                ],
-                [
-                  665,
-                  671
-                ],
-                [
-                  931,
-                  937
-                ],
-                [
-                  1843,
-                  1849
-                ],
-                [
-                  1868,
-                  1874
-                ]
-              ],
-              "right_ranges": [
-                [
-                  54,
-                  61
-                ],
-                [
-                  587,
-                  593
-                ]
-              ]
-            }
-          ]
-        },
-        {
-          "left_pr_number": 43836,
-          "right_pr_number": 43842,
-          "code_similarity": 0.332,
-          "size_similarity": 0.088,
-          "file_overlap": 1.0,
-          "area_overlap": 0.033,
-          "patch_similarity": 0.017,
-          "shared_filenames": [
-            "src/transformers/cli/serve.py"
-          ],
-          "shared_file_areas": [
-            {
-              "filename": "src/transformers/cli/serve.py",
-              "left_ranges": [
-                [
-                  11,
-                  18
-                ],
-                [
-                  359,
-                  429
-                ],
-                [
-                  584,
-                  590
-                ],
-                [
-                  1892,
-                  1910
-                ],
-                [
-                  1917,
-                  1923
-                ]
-              ],
-              "right_ranges": [
-                [
-                  54,
-                  61
-                ],
-                [
-                  587,
-                  593
-                ]
-              ]
-            }
-          ]
-        }
-      ]
-    },
     {
       "cluster_id": "cluster-43240-3",
       "summary": "Cluster of 1 issues and 2 PRs centered on issue #43240.",
@@ -1620,6 +1422,76 @@
         }
       ]
     },
+    {
+      "cluster_id": "cluster-43824-3",
+      "summary": "Cluster of 1 issues and 2 PRs centered on issue #43824.",
+      "status": "open",
+      "confidence": 0.75,
+      "canonical_issue_number": 43824,
+      "canonical_pr_number": 43836,
+      "issue_numbers": [
+        43824
+      ],
+      "pr_numbers": [
+        43836,
+        43842
+      ],
+      "evidence_types": [
+        "closing_reference",
+        "shared_issue_target"
+      ],
+      "pr_comparisons": [
+        {
+          "left_pr_number": 43836,
+          "right_pr_number": 43842,
+          "code_similarity": 0.332,
+          "size_similarity": 0.088,
+          "file_overlap": 1.0,
+          "area_overlap": 0.033,
+          "patch_similarity": 0.017,
+          "shared_filenames": [
+            "src/transformers/cli/serve.py"
+          ],
+          "shared_file_areas": [
+            {
+              "filename": "src/transformers/cli/serve.py",
+              "left_ranges": [
+                [
+                  11,
+                  18
+                ],
+                [
+                  359,
+                  429
+                ],
+                [
+                  584,
+                  590
+                ],
+                [
+                  1892,
+                  1910
+                ],
+                [
+                  1917,
+                  1923
+                ]
+              ],
+              "right_ranges": [
+                [
+                  54,
+                  61
+                ],
+                [
+                  587,
+                  593
+                ]
+              ]
+            }
+          ]
+        }
+      ]
+    },
     {
       "cluster_id": "cluster-45081-3",
       "summary": "Cluster of 1 issues and 2 PRs centered on issue #45081.",
@@ -1813,54 +1685,277 @@
       ]
     },
     {
-      "cluster_id": "cluster-44002-3",
-      "summary": "Cluster of 3 related pull requests linked by soft_similarity.",
+      "cluster_id": "cluster-44018-2",
+      "summary": "Cluster of 2 related pull requests linked by soft_similarity.",
       "status": "open",
       "confidence": 0.5,
       "canonical_issue_number": null,
-      "canonical_pr_number": 44002,
+      "canonical_pr_number": 44068,
       "issue_numbers": [],
       "pr_numbers": [
-        44002,
-        44004,
-        44029
+        44018,
+        44068
       ],
       "evidence_types": [
         "soft_similarity"
       ],
       "pr_comparisons": [
         {
-          "left_pr_number": 44002,
-          "right_pr_number": 44004,
-          "code_similarity": 0.125,
-          "size_similarity": 0.398,
-          "file_overlap": 0.0,
-          "area_overlap": 0.0,
-          "patch_similarity": 0.304,
-          "shared_filenames": [],
-          "shared_file_areas": []
-        },
-        {
-          "left_pr_number": 44002,
-          "right_pr_number": 44029,
-          "code_similarity": 0.135,
-          "size_similarity": 0.43,
-          "file_overlap": 0.0,
-          "area_overlap": 0.0,
-          "patch_similarity": 0.329,
-          "shared_filenames": [],
-          "shared_file_areas": []
-        },
-        {
-          "left_pr_number": 44004,
-          "right_pr_number": 44029,
-          "code_similarity": 0.253,
-          "size_similarity": 0.925,
-          "file_overlap": 0.0,
-          "area_overlap": 0.0,
-          "patch_similarity": 0.456,
-          "shared_filenames": [],
-          "shared_file_areas": []
+          "left_pr_number": 44018,
+          "right_pr_number": 44068,
+          "code_similarity": 0.766,
+          "size_similarity": 0.939,
+          "file_overlap": 1.0,
+          "area_overlap": 0.425,
+          "patch_similarity": 0.866,
+          "shared_filenames": [
+            "src/transformers/models/gpt_neo/modeling_gpt_neo.py",
+            "tests/models/gpt_neo/test_modeling_gpt_neo.py"
+          ],
+          "shared_file_areas": [
+            {
+              "filename": "src/transformers/models/gpt_neo/modeling_gpt_neo.py",
+              "left_ranges": [
+                [
+                  26,
+                  46
+                ],
+                [
+                  138,
+                  143
+                ],
+                [
+                  182,
+                  187
+                ],
+                [
+                  283,
+                  295
+                ],
+                [
+                  327,
+                  341
+                ],
+                [
+                  348,
+                  354
+                ],
+                [
+                  360,
+                  369
+                ],
+                [
+                  399,
+                  406
+                ],
+                [
+                  411,
+                  419
+                ],
+                [
+                  428,
+                  433
+                ],
+                [
+                  472,
+                  492
+                ],
+                [
+                  507,
+                  513
+                ],
+                [
+                  519,
+                  528
+                ],
+                [
+                  541,
+                  559
+                ],
+                [
+                  562,
+                  567
+                ],
+                [
+                  595,
+                  601
+                ],
+                [
+                  607,
+                  614
+                ],
+                [
+                  627,
+                  643
+                ],
+                [
+                  685,
+                  690
+                ],
+                [
+                  708,
+                  714
+                ],
+                [
+                  720,
+                  727
+                ],
+                [
+                  740,
+                  757
+                ],
+                [
+                  761,
+                  766
+                ],
+                [
+                  780,
+                  786
+                ],
+                [
+                  791,
+                  798
+                ],
+                [
+                  807,
+                  822
+                ],
+                [
+                  840,
+                  845
+                ]
+              ],
+              "right_ranges": [
+                [
+                  26,
+                  31
+                ],
+                [
+                  34,
+                  44
+                ],
+                [
+                  136,
+                  141
+                ],
+                [
+                  180,
+                  185
+                ],
+                [
+                  281,
+                  293
+                ],
+                [
+                  325,
+                  339
+                ],
+                [
+                  346,
+                  352
+                ],
+                [
+                  358,
+                  367
+                ],
+                [
+                  397,
+                  404
+                ],
+                [
+                  409,
+                  417
+                ],
+                [
+                  426,
+                  434
+                ],
+                [
+                  463,
+                  483
+                ],
+                [
+                  498,
+                  504
+                ],
+                [
+                  510,
+                  519
+                ],
+                [
+                  532,
+                  550
+                ],
+                [
+                  553,
+                  564
+                ],
+                [
+                  586,
+                  592
+                ],
+                [
+                  598,
+                  605
+                ],
+                [
+                  618,
+                  634
+                ],
+                [
+                  676,
+                  688
+                ],
+                [
+                  699,
+                  705
+                ],
+                [
+                  711,
+                  718
+                ],
+                [
+                  731,
+                  748
+                ],
+                [
+                  752,
+                  762
+                ],
+                [
+                  771,
+                  777
+                ],
+                [
+                  782,
+                  789
+                ],
+                [
+                  798,
+                  813
+                ],
+                [
+                  831,
+                  836
+                ]
+              ]
+            },
+            {
+              "filename": "tests/models/gpt_neo/test_modeling_gpt_neo.py",
+              "left_ranges": [
+                [
+                  458,
+                  464
+                ]
+              ],
+              "right_ranges": [
+                [
+                  458,
+                  464
+                ]
+              ]
+            }
+          ]
         }
       ]
     }
@@ -1894,16 +1989,6 @@
       "target_issue_number": 43240,
       "reason": "PRs in cluster-43240-3 are treated as duplicates because they converge on issue #43240 with closing_reference, shared_issue_target evidence."
     },
-    {
-      "cluster_id": "cluster-43656-4",
-      "canonical_pr_number": 43656,
-      "duplicate_pr_numbers": [
-        43836,
-        43842
-      ],
-      "target_issue_number": 43824,
-      "reason": "PRs in cluster-43656-4 are treated as duplicates because they converge on issue #43824 with closing_reference, shared_issue_target, soft_similarity evidence."
-    },
     {
       "cluster_id": "cluster-43698-3",
       "canonical_pr_number": 43779,
@@ -1913,6 +1998,15 @@
       "target_issue_number": 43698,
       "reason": "PRs in cluster-43698-3 are treated as duplicates because they converge on issue #43698 with closing_reference, shared_issue_target evidence."
     },
+    {
+      "cluster_id": "cluster-43824-3",
+      "canonical_pr_number": 43836,
+      "duplicate_pr_numbers": [
+        43842
+      ],
+      "target_issue_number": 43824,
+      "reason": "PRs in cluster-43824-3 are treated as duplicates because they converge on issue #43824 with closing_reference, shared_issue_target evidence."
+    },
     {
       "cluster_id": "cluster-43979-11",
       "canonical_pr_number": 44007,
@@ -1931,14 +2025,13 @@
       "reason": "PRs in cluster-43979-11 are treated as duplicates because they converge on issue #43979 with closing_reference, shared_issue_target, soft_similarity evidence."
     },
     {
-      "cluster_id": "cluster-44002-3",
-      "canonical_pr_number": 44002,
+      "cluster_id": "cluster-44018-2",
+      "canonical_pr_number": 44068,
       "duplicate_pr_numbers": [
-        44004,
-        44029
+        44018
       ],
       "target_issue_number": null,
-      "reason": "PRs in cluster-44002-3 are treated as duplicates because they share soft_similarity evidence."
+      "reason": "PRs in cluster-44018-2 are treated as duplicates because they share soft_similarity evidence."
     },
     {
       "cluster_id": "cluster-45081-3",
diff --git a/analysis/current/analysis-report-hybrid.llm-reviews.json b/analysis/current/analysis-report-hybrid.llm-reviews.json
index 229dab992f065c96463b59680e2b1bd7f0f2c848..95c28fd896e0d289d7282bd98dee50b12a8147e2 100644
--- a/analysis/current/analysis-report-hybrid.llm-reviews.json
+++ b/analysis/current/analysis-report-hybrid.llm-reviews.json
@@ -1,8 +1,8 @@
 {
   "schema_version": "1.0",
   "repo": "huggingface/transformers",
-  "snapshot_id": "20260510T000023Z",
-  "generated_at": "2026-05-10T00:05:56Z",
+  "snapshot_id": "20260510T060019Z",
+  "generated_at": "2026-05-10T06:03:17Z",
   "model": "gpt-5.4-mini?service_tier=flex",
   "reviews": [
     {
@@ -13,8 +13,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-13244-3",
       "nodes": [
@@ -84,8 +84,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -632,8 +632,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -975,8 +975,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -1325,8 +1325,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -1653,8 +1653,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -1974,8 +1974,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -2317,8 +2317,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -2643,8 +2643,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -2997,8 +2997,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -3308,8 +3308,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -3641,8 +3641,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -3966,8 +3966,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -4281,8 +4281,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -4578,8 +4578,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -4883,8 +4883,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -5174,8 +5174,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -5485,8 +5485,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -5813,8 +5813,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -6096,8 +6096,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -6428,8 +6428,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -6761,8 +6761,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -7055,8 +7055,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -7387,8 +7387,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -7737,8 +7737,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -8042,8 +8042,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -8347,8 +8347,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -8645,8 +8645,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -8936,8 +8936,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -9255,8 +9255,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -9559,8 +9559,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -9872,8 +9872,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -10177,8 +10177,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -10474,8 +10474,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -10785,8 +10785,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -11108,8 +11108,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -11430,8 +11430,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -11765,8 +11765,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -12076,8 +12076,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -12416,8 +12416,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -12721,8 +12721,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -13076,8 +13076,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -13422,8 +13422,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -13761,8 +13761,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -14115,8 +14115,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -14491,8 +14491,8 @@
       "review_unit_count": 45,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-15354-392",
       "nodes": [
@@ -14932,8 +14932,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-26350-2",
       "nodes": [
@@ -14995,8 +14995,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-27088-4",
       "nodes": [
@@ -15095,8 +15095,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-36817-15",
       "nodes": [
@@ -15409,8 +15409,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-37934-6",
       "nodes": [
@@ -15525,8 +15525,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-41084-2",
       "nodes": [
@@ -15588,8 +15588,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-43598-2",
       "nodes": [
@@ -15651,8 +15651,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-43684-2",
       "nodes": [
@@ -15714,8 +15714,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-43835-4",
       "nodes": [
@@ -15814,8 +15814,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-44371-5",
       "nodes": [
@@ -15936,8 +15936,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-45145-2",
       "nodes": [
@@ -16056,8 +16056,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-45672-2",
       "nodes": [
@@ -16119,8 +16119,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-45762-2",
       "nodes": [
@@ -16182,8 +16182,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "llm",
-      "cache_hit": false,
+      "source": "cache",
+      "cache_hit": true,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-45764-2",
       "nodes": [
@@ -16245,8 +16245,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "cache",
-      "cache_hit": true,
+      "source": "llm",
+      "cache_hit": false,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-39895-13",
       "nodes": [
@@ -16266,8 +16266,8 @@
       ],
       "soft_pairs": [
         "pull_request:39895|pull_request:43424",
-        "pull_request:41855|pull_request:43823",
         "pull_request:41855|pull_request:43424",
+        "pull_request:41855|pull_request:43823",
         "pull_request:42668|pull_request:43823",
         "pull_request:43251|pull_request:43823",
         "pull_request:43251|pull_request:44827",
@@ -16285,12 +16285,12 @@
         "pull_request:43823|pull_request:45795",
         "pull_request:43251|pull_request:43254"
       ],
-      "prepared_review_unit_hash": "96bf7125b236f5dc8d123aa48d42745a15b3df7129e3b65b913ee207e74cfe35",
+      "prepared_review_unit_hash": "89c73c8ba9e08bb158f968e0590ca6071f3427f941b3b30a742a310d6c241f26",
       "component_budget": {
         "node_count": 13,
         "item_count": 13,
         "soft_pair_count": 19,
-        "serialized_chars": 35205,
+        "serialized_chars": 35208,
         "estimated_input_tokens": 8802,
         "estimated_eval_tokens": 17860
       },
@@ -16298,7 +16298,7 @@
         "node_count": 13,
         "item_count": 13,
         "soft_pair_count": 19,
-        "serialized_chars": 35205,
+        "serialized_chars": 35208,
         "estimated_input_tokens": 8802,
         "estimated_eval_tokens": 17860
       },
@@ -16307,135 +16307,135 @@
       "aggressively_trimmed": false,
       "split": false,
       "analyst_result": {
-        "summary": "This set is mostly a grab bag of unrelated PRs. The only true duplicate-like pair is the two `fixed_cross_entropy` patches (43251 and 43254), which target the same issue and same code path; all other soft-similarity pairs are too broad or clearly different changes.",
-        "confidence": 0.98,
+        "summary": "Mostly unrelated PRs with one real duplicate-like pair in `fixed_cross_entropy`. The loss-helper pair (43251/43254) is the only strong merge candidate; the rest are different models, docs, tests, or subsystems.",
+        "confidence": 0.94,
         "canonical_issue_reason": null,
-        "canonical_pr_reason": "PR 43251 is the best canonical representative for the cross-entropy kwargs fix: it directly targets issue 43240, patches the shared helper in the intended way, and is the clearer of the two near-identical variants.",
+        "canonical_pr_reason": "PR 43251 is a reasonable canonical representative for the loss-helper pair because it makes the same `fixed_cross_entropy` change in `src/transformers/loss/loss_utils.py` as PR 43254 and both target issue 43240. The packet does not clearly justify a stronger preference beyond that.",
         "best_issue_reason": null,
-        "best_pr_reason": "PR 43251 is the strongest standalone PR in the only real duplicate pair; it is a focused bugfix with an explicit issue link and broader, cleaner coverage than 43254.",
+        "best_pr_reason": "PR 43251 is a suitable global representative only for the one duplicate-like fix in this cluster: it and PR 43254 patch the same helper with the same supported kwargs in the same file, so either could stand in for that pair.",
         "soft_edge_verdicts": [
           {
             "left": "pull_request:39895",
             "right": "pull_request:43424",
             "accept": false,
-            "reason": "Different changes: VideoPrism model addition vs. an ExecuTorch dynamic-shape test."
+            "reason": "VideoPrism integration vs an ExecuTorch dynamic-shape test; different changes and no shared concrete bug."
           },
           {
             "left": "pull_request:41855",
-            "right": "pull_request:43823",
+            "right": "pull_request:43424",
             "accept": false,
-            "reason": "Unrelated feature areas: Mistral tokenizer API methods vs. adding MobileLLM."
+            "reason": "Mistral tokenizer helper additions vs ExecuTorch export testing; unrelated code paths."
           },
           {
             "left": "pull_request:41855",
-            "right": "pull_request:43424",
+            "right": "pull_request:43823",
             "accept": false,
-            "reason": "Tokenizer method fix and ExecuTorch export test are not the same bug/change."
+            "reason": "Tokenizer method fixes vs adding MobileLLM; same broad area only, not the same fix."
           },
           {
             "left": "pull_request:42668",
             "right": "pull_request:43823",
             "accept": false,
-            "reason": "Both touch pretrained/processor loading only in a very broad sense; they fix different models and different problems."
+            "reason": "Processor/auto-registration work for different models; not the same underlying change."
           },
           {
             "left": "pull_request:43251",
             "right": "pull_request:43823",
             "accept": false,
-            "reason": "Cross-entropy helper kwargs fix is unrelated to adding MobileLLM."
+            "reason": "Loss helper kwargs fix vs new model implementation; no common code-path problem."
           },
           {
             "left": "pull_request:43251",
             "right": "pull_request:44827",
             "accept": false,
-            "reason": "Different targets: loss helper kwargs vs. Mistral4 test/model adjustments."
+            "reason": "Cross-entropy helper fix vs Mistral4 model/test changes; different bugs."
           },
           {
             "left": "pull_request:43424",
             "right": "pull_request:44827",
             "accept": false,
-            "reason": "ExecuTorch dynamic-shape test does not match Mistral4 test fixes."
+            "reason": "ExecuTorch dynamic-shape test vs Mistral4 fixes; same general testing theme only."
           },
           {
             "left": "pull_request:43424",
             "right": "pull_request:43823",
             "accept": false,
-            "reason": "Exportability test and new model implementation are unrelated."
+            "reason": "ExecuTorch export test vs MobileLLM addition; unrelated."
           },
           {
             "left": "pull_request:43823",
             "right": "pull_request:45655",
             "accept": false,
-            "reason": "Model addition and config-resolution fix are distinct changes."
+            "reason": "New model addition vs `cls.config` resolution fix; different subsystems."
           },
           {
             "left": "pull_request:43823",
             "right": "pull_request:43995",
             "accept": false,
-            "reason": "Adding MobileLLM is unrelated to Falcon output-interface refactoring."
+            "reason": "MobileLLM implementation vs Falcon refactor; not duplicates."
           },
           {
             "left": "pull_request:43995",
             "right": "pull_request:44827",
             "accept": false,
-            "reason": "Different models and different code paths; no shared concrete bug."
+            "reason": "Falcon output-collection refactor vs Mistral4 test/model work; different models and goals."
           },
           {
             "left": "pull_request:43995",
             "right": "pull_request:45213",
             "accept": false,
-            "reason": "Falcon refactor is unrelated to the model-creation skill PR."
+            "reason": "Falcon refactor vs a model-creation skill PR; completely different artifact types."
           },
           {
             "left": "pull_request:44827",
             "right": "pull_request:45170",
             "accept": false,
-            "reason": "Mistral4 test fixes and `layrnorm` renaming are separate issues."
+            "reason": "Mistral4 fixes vs CLIP-like `layrnorm` renames; different underlying issues."
           },
           {
             "left": "pull_request:44827",
             "right": "pull_request:45795",
             "accept": false,
-            "reason": "Mistral4 testing/model fixes are unrelated to docs for audio/video processors."
+            "reason": "Mistral4 work vs audio/video processor docs; not the same change."
           },
           {
             "left": "pull_request:44827",
             "right": "pull_request:45655",
             "accept": false,
-            "reason": "Model test fixes do not match the `cls.config` resolution order bug."
+            "reason": "Model-specific fixes vs core config-resolution ordering; no shared code-path bug."
           },
           {
             "left": "pull_request:44827",
             "right": "pull_request:45213",
             "accept": false,
-            "reason": "Shared auto/modeling filenames are incidental; one is a model PR, the other is a docs/skill scaffold."
+            "reason": "Model-specific fixes vs repo skill/dataset PR; unrelated."
           },
           {
             "left": "pull_request:43823",
             "right": "pull_request:45213",
             "accept": false,
-            "reason": "Adding MobileLLM is not the same as adding the model-creation skill docs."
+            "reason": "New model implementation vs draft model-creation skill/docs; same domain only."
           },
           {
             "left": "pull_request:43823",
             "right": "pull_request:45795",
             "accept": false,
-            "reason": "MobileLLM model code and processor docs are different artifacts."
+            "reason": "Model implementation vs docs for audio/video processors; related topic, not a duplicate."
           },
           {
             "left": "pull_request:43251",
             "right": "pull_request:43254",
             "accept": true,
-            "reason": "Same underlying bug in `fixed_cross_entropy`, same shared helper, same issue target 43240; these are duplicate implementation variants."
+            "reason": "Both PRs change `src/transformers/loss/loss_utils.py` in the same `fixed_cross_entropy` helper, add the same kwargs (`weight`, `label_smoothing`), and target the same issue 43240. They are the same underlying fix and could be merged into one PR."
           }
         ]
       },
       "evaluator_result": {
-        "accept": true,
-        "feedback": "The assessment is well-grounded in the packet. All non-43251/43254 pairs are conservatively rejected as only soft-similar or clearly different changes, and 43251/43254 are correctly identified as the only mergeable same-bugfix pair because they touch the same helper in the same file and share issue target 43240. No evident overstatement beyond the packet."
+        "accept": false,
+        "feedback": "Reject: the canonical PR rationale overstates the packet. PR 43254 also has explicit issue target 43240, so it does not \"lack the issue linkage.\" The claim that 43251 is broader/cleaner is also not directly grounded in the packet. The accepted duplicate pair itself looks mergeable, but the summary should be more conservative."
       },
       "evaluator_used": true,
-      "retried": false,
+      "retried": true,
       "accepted_nontrivial_soft_edge": true,
       "error_kind": null,
       "error_message": null
@@ -16448,8 +16448,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "cache",
-      "cache_hit": true,
+      "source": "llm",
+      "cache_hit": false,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-40546-2",
       "nodes": [
@@ -16459,7 +16459,7 @@
       "soft_pairs": [
         "pull_request:40546|pull_request:42781"
       ],
-      "prepared_review_unit_hash": "78353c94123f971b64c3f40c43ca133ce1fdff7b4c583072a5f85afc263a2e93",
+      "prepared_review_unit_hash": "582e4f4f7ae86a8f5aa999749fca767e875d77206a05f6c275d78d4defc91332",
       "component_budget": {
         "node_count": 2,
         "item_count": 2,
@@ -16481,18 +16481,18 @@
       "aggressively_trimmed": false,
       "split": false,
       "analyst_result": {
-        "summary": "Both items are VibeVoice implementation PRs, but they are not the same concrete change: #40546 is the base VibeVoice integration, while #42781 adds a separate realtime variant plus an acoustic tokenizer and related generation/model updates. They overlap heavily by file path because #42781 extends the same subsystem, but it is not a duplicate merge candidate of the earlier PR.",
-        "confidence": 0.83,
+        "summary": "Both PRs are about VibeVoice, but they target different deliverables: one adds the base VibeVoice model integration, while the other adds VibeVoice Realtime and an acoustic tokenizer with additional generation/runtime changes. They overlap on shared model scaffolding, but not on the same concrete change.",
+        "confidence": 0.93,
         "canonical_issue_reason": null,
-        "canonical_pr_reason": "#42781 is the broader and more up-to-date VibeVoice PR, covering the realtime path and the acoustic tokenizer alongside the core model integration.",
+        "canonical_pr_reason": "PR #40546 is the better canonical PR because it is the foundational VibeVoice implementation and broader base integration; the other PR builds a separate realtime variant on top of it.",
         "best_issue_reason": null,
-        "best_pr_reason": "#42781 is the best representative of the cluster because it contains the later, more complete VibeVoice work and subsumes the base implementation context.",
+        "best_pr_reason": "PR #40546 is the best representative of the cluster because it covers the core model addition and is the more central, non-draft implementation.",
         "soft_edge_verdicts": [
           {
             "left": "pull_request:40546",
             "right": "pull_request:42781",
             "accept": false,
-            "reason": "Shared filenames reflect a common VibeVoice subsystem, but the change sets are not the same underlying feature: #40546 implements the base model, while #42781 adds realtime support and an acoustic tokenizer. They are related follow-up work, not duplicate PRs."
+            "reason": "Reject: they share the VibeVoice family and some files, but one is the base model integration and the other is a realtime/acoustic-tokenizer extension with different code paths and scope. They do not look like the same mergeable fix/change."
           }
         ]
       },
@@ -16511,8 +16511,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "cache",
-      "cache_hit": true,
+      "source": "llm",
+      "cache_hit": false,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-40755-2",
       "nodes": [
@@ -16522,12 +16522,12 @@
       "soft_pairs": [
         "pull_request:40755|pull_request:41224"
       ],
-      "prepared_review_unit_hash": "c42c80c881d0c96ca69b5537803885ad027cee853e1d8f5c325054a812323847",
+      "prepared_review_unit_hash": "288f99592cd9551e0c534578064187a69297a2b7027bcd7d01c25f029b5f9df3",
       "component_budget": {
         "node_count": 2,
         "item_count": 2,
         "soft_pair_count": 1,
-        "serialized_chars": 4674,
+        "serialized_chars": 4676,
         "estimated_input_tokens": 1169,
         "estimated_eval_tokens": 2594
       },
@@ -16535,7 +16535,7 @@
         "node_count": 2,
         "item_count": 2,
         "soft_pair_count": 1,
-        "serialized_chars": 4674,
+        "serialized_chars": 4676,
         "estimated_input_tokens": 1169,
         "estimated_eval_tokens": 2594
       },
@@ -16544,18 +16544,18 @@
       "aggressively_trimmed": false,
       "split": false,
       "analyst_result": {
-        "summary": "The two pull requests are unrelated feature additions in different model families (TimesFM covariates vs. DINOv3 image classification), so this is not a duplicate cluster.",
+        "summary": "The two pull requests are unrelated: one adds TimesFM covariate forecasting, the other adds DINOv3 image classification support. They touch different models, code paths, and documentation, so this is not a duplicate cluster.",
         "confidence": 0.99,
         "canonical_issue_reason": null,
         "canonical_pr_reason": null,
         "best_issue_reason": null,
-        "best_pr_reason": "No single PR is a good canonical representative because the items address different models and different changes; any choice would be arbitrary.",
+        "best_pr_reason": "PR #40755 is the broader, more substantial feature addition in this cluster, but it is not a duplicate of #41224 and should not be treated as a canonical match.",
         "soft_edge_verdicts": [
           {
             "left": "pull_request:40755",
             "right": "pull_request:41224",
             "accept": false,
-            "reason": "Different underlying features for different architectures and code paths; they do not look mergeable into one PR or like the same bug/change."
+            "reason": "Different models and features (TimesFM covariates vs. DINOv3 image classification); no shared underlying bug or mergeable change."
           }
         ]
       },
@@ -16637,8 +16637,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "cache",
-      "cache_hit": true,
+      "source": "llm",
+      "cache_hit": false,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-41797-3",
       "nodes": [
@@ -16650,7 +16650,7 @@
         "pull_request:41797|pull_request:45186",
         "pull_request:45186|pull_request:45630"
       ],
-      "prepared_review_unit_hash": "c9907a1ec37f09a7f804609844aa4b0a2c8af30343572614eef6507d28ebefd8",
+      "prepared_review_unit_hash": "742e11c30f1d1eb9c7d3b835a1b501e3ea127e469d11f7e9abdad35eda54cf1f",
       "component_budget": {
         "node_count": 3,
         "item_count": 3,
@@ -16672,24 +16672,24 @@
       "aggressively_trimmed": false,
       "split": false,
       "analyst_result": {
-        "summary": "These are three separate model-addition PRs (DeepSeek OCR, Isaac, and Kimi2-6). They share some Hugging Face integration scaffolding, but they target different models, different files, and different code paths, so they are not duplicates.",
-        "confidence": 0.98,
+        "summary": "These are three separate model-addition PRs (DeepSeek OCR, Isaac, Kimi2-6). They share generic Transformers boilerplate and registry/docs updates, but not the same underlying model or code-path change.",
+        "confidence": 0.97,
         "canonical_issue_reason": null,
-        "canonical_pr_reason": "No canonical PR for duplicate triage: each PR adds a different model package and associated integrations, so there is no single underlying change to collapse them into.",
+        "canonical_pr_reason": "No canonical PR: the cluster is heterogeneous, with each PR introducing a different model integration rather than duplicating the same change.",
         "best_issue_reason": null,
-        "best_pr_reason": "No global best PR for deduplication: the PRs are unrelated feature additions rather than alternate fixes for the same bug/change.",
+        "best_pr_reason": "PR 45630 is the most developed integration by discussion/review activity and scope, so it is the best representative of the cluster, but it is still not a duplicate of the others.",
         "soft_edge_verdicts": [
           {
             "left": "pull_request:41797",
             "right": "pull_request:45186",
             "accept": false,
-            "reason": "Different model additions (DeepSeek OCR vs Isaac) with distinct model directories and docs; shared auto-registration files are generic scaffolding, not evidence of the same change."
+            "reason": "Different model integrations (DeepSeek OCR vs Isaac). The overlap is limited to standard new-model scaffolding in docs, auto mappings, and utility updates, not the same bug or feature."
           },
           {
             "left": "pull_request:45186",
             "right": "pull_request:45630",
             "accept": false,
-            "reason": "Isaac vs Kimi2-6 are separate model integrations with different filenames and model code. Similar Hugging Face wiring is expected for new-model PRs, but they do not fix or add the same concrete feature."
+            "reason": "Different model integrations (Isaac vs Kimi2-6). Shared registry/docs churn is expected for new models and does not indicate a common underlying code-path change."
           }
         ]
       },
@@ -16708,8 +16708,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "cache",
-      "cache_hit": true,
+      "source": "llm",
+      "cache_hit": false,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-42975-2",
       "nodes": [
@@ -16719,7 +16719,7 @@
       "soft_pairs": [
         "pull_request:42975|pull_request:42976"
       ],
-      "prepared_review_unit_hash": "b09fb122c53437bed9b33a659e1a40ddb1f31b3611b1037754be5531dc1a010d",
+      "prepared_review_unit_hash": "a3fc123a430a9441552d609af6d627312dbe0c7aba616f0b8d026cc87d346c2a",
       "component_budget": {
         "node_count": 2,
         "item_count": 2,
@@ -16741,18 +16741,18 @@
       "aggressively_trimmed": false,
       "split": false,
       "analyst_result": {
-        "summary": "Two open PRs both update GitHub Actions dependencies in workflow files, but they target different actions and different maintenance goals: one is a broad Node 24 compatibility sweep, the other is specifically bumping docker/build-push-action to v6. They are related at the workflow-maintenance level, but not the same change.",
+        "summary": "Two workflow-maintenance PRs that both upgrade GitHub Actions, but they target different actions and different goals. They are related by theme, not the same underlying change, so they should not be deduplicated.",
         "confidence": 0.95,
         "canonical_issue_reason": null,
-        "canonical_pr_reason": "PR 42975 is the better canonical PR because its scope is broader and its title clearly describes the main maintenance intent: upgrading GitHub Actions for Node 24 compatibility across many workflows.",
+        "canonical_pr_reason": "No duplicate canonical PR: the two PRs address different action upgrades and are not the same mergeable change.",
         "best_issue_reason": null,
-        "best_pr_reason": "PR 42975 is the best representative of the cluster since it covers the broader and more central workflow-action upgrade theme; PR 42976 is a narrower, action-specific version bump.",
+        "best_pr_reason": "PR 42975 is the better representative of the cluster because it is the broader workflow-actions compatibility update (checkout/cache/upload-artifact) and explicitly targets Node 24 compatibility.",
         "soft_edge_verdicts": [
           {
             "left": "pull_request:42975",
             "right": "pull_request:42976",
             "accept": false,
-            "reason": "Both are workflow dependency upgrades, but they change different actions and address different problems: 42975 updates multiple actions for Node 24 compatibility, while 42976 only bumps docker/build-push-action. They are not the same underlying fix and would not plausibly be merged as one PR."
+            "reason": "Both modify GitHub Actions workflows, but 42975 upgrades checkout/cache/upload-artifact for Node 24 compatibility while 42976 bumps docker/build-push-action to v6. Different actions, different concrete change, not the same underlying bug/fix."
           }
         ]
       },
@@ -16771,8 +16771,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "cache",
-      "cache_hit": true,
+      "source": "llm",
+      "cache_hit": false,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-42978-7",
       "nodes": [
@@ -16795,88 +16795,88 @@
         "pull_request:44601|pull_request:45838",
         "pull_request:45490|pull_request:45838"
       ],
-      "prepared_review_unit_hash": "7f92cac352ce3feca71e086a66b59b42f07bc2616fb7895c7dfca6d931a08883",
+      "prepared_review_unit_hash": "ced02bef74514f2010a2792937efbb35b97d750f38f48d45a37a624086090049",
       "component_budget": {
         "node_count": 7,
         "item_count": 7,
         "soft_pair_count": 9,
-        "serialized_chars": 23511,
-        "estimated_input_tokens": 5878,
-        "estimated_eval_tokens": 12012
+        "serialized_chars": 23514,
+        "estimated_input_tokens": 5879,
+        "estimated_eval_tokens": 12014
       },
       "budget": {
         "node_count": 7,
         "item_count": 7,
         "soft_pair_count": 9,
-        "serialized_chars": 23511,
-        "estimated_input_tokens": 5878,
-        "estimated_eval_tokens": 12012
+        "serialized_chars": 23514,
+        "estimated_input_tokens": 5879,
+        "estimated_eval_tokens": 12014
       },
       "overflow_policy": "truncate_then_skip",
       "trimmed": false,
       "aggressively_trimmed": false,
       "split": false,
       "analyst_result": {
-        "summary": "This is a noisy cluster of unrelated PRs that mostly happen to add new models or shared registration/docs changes. The overlapping filenames are generic auto-mapping/documentation plumbing, not evidence of the same underlying bug or feature.",
-        "confidence": 0.95,
+        "summary": "These PRs are mostly unrelated model additions or infrastructure work; the high soft-similarity scores come from shared Hugging Face boilerplate, docs, and auto-registration files rather than the same underlying change. I found no true duplicate PR pair in this cluster.",
+        "confidence": 0.97,
         "canonical_issue_reason": null,
-        "canonical_pr_reason": "If a single anchor is needed, PR 45838 is the broadest self-contained model-addition change in the set, with multiple related model files plus tests/docs. That said, the cluster is heterogeneous and does not form a true duplicate set.",
+        "canonical_pr_reason": "No canonical PR for dedupe: the cluster spans separate changes for ViT NEPA, Molmo, Molmo2, LFM2 audio, distributed pipeline-parallel loading, CTSM, and PP-OCRv6, with only template/docs overlap.",
         "best_issue_reason": null,
-        "best_pr_reason": "PR 45838 is the most representative anchor only because it is a larger, fully fleshed-out model-support PR; it is not a better duplicate candidate than the others, just the most substantial single change in the cluster.",
+        "best_pr_reason": "No PR is a strong canonical duplicate target here; if a representative must be chosen, 43451 is the most substantial standalone model-addition PR, but it is not a duplicate of the others.",
         "soft_edge_verdicts": [
           {
             "left": "pull_request:42978",
             "right": "pull_request:43448",
             "accept": false,
-            "reason": "Different models: ViT NEPA vs Molmo. Shared docs/auto-registration files are generic scaffolding, not the same change."
+            "reason": "Different model additions (ViT NEPA vs Molmo); overlap is only in docs and auto-mapping boilerplate."
           },
           {
             "left": "pull_request:42978",
             "right": "pull_request:43451",
             "accept": false,
-            "reason": "Different model families: ViT NEPA vs Molmo2. Overlap is only in common registration/docs plumbing."
+            "reason": "Different model families (ViT NEPA vs Molmo2) and distinct implementation paths; shared files are generic registration/docs."
           },
           {
             "left": "pull_request:42978",
             "right": "pull_request:45490",
             "accept": false,
-            "reason": "Different concrete additions: ViT NEPA vs CTSM. No shared bug or code path."
+            "reason": "ViT NEPA model addition and CTSM time-series model addition are unrelated changes."
           },
           {
             "left": "pull_request:43448",
             "right": "pull_request:43451",
             "accept": false,
-            "reason": "Molmo and Molmo2 are distinct model lines with different implementations; not the same PR-worthy change."
+            "reason": "Molmo and Molmo2 are related by name but are separate model releases with different code paths, not the same bug/change."
           },
           {
             "left": "pull_request:43448",
             "right": "pull_request:45490",
             "accept": false,
-            "reason": "Molmo vs CTSM are unrelated model additions; shared framework files are too broad to indicate duplication."
+            "reason": "Molmo vs CTSM are unrelated model additions; only broad docs/auto files overlap."
           },
           {
             "left": "pull_request:43451",
             "right": "pull_request:45490",
             "accept": false,
-            "reason": "Molmo2 and CTSM are unrelated model families; the overlap is only generic model registration."
+            "reason": "Molmo2 and CTSM touch different subsystems and model implementations, so they are not duplicates."
           },
           {
             "left": "pull_request:43973",
             "right": "pull_request:45838",
             "accept": false,
-            "reason": "Audio model support vs OCR text-recognition models. Different tasks, model code paths, and test surfaces."
+            "reason": "Lfm2 audio support and PP-OCRv6 text recognition support are unrelated model families with different code paths."
           },
           {
             "left": "pull_request:44601",
             "right": "pull_request:45838",
             "accept": false,
-            "reason": "Pipeline-parallel loading infrastructure is a separate cross-cutting feature, not the same change as PP-OCRv6 model support."
+            "reason": "Distributed pipeline-parallel loading infrastructure is unrelated to PP-OCRv6 model support; shared similarity is incidental."
           },
           {
             "left": "pull_request:45490",
             "right": "pull_request:45838",
             "accept": false,
-            "reason": "CTSM and PP-OCRv6/PPLCNetV4 are different model families; shared auto/docs edits do not make them duplicates."
+            "reason": "CTSM time-series model support and PP-OCRv6 OCR model support are unrelated changes."
           }
         ]
       },
@@ -16895,8 +16895,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "cache",
-      "cache_hit": true,
+      "source": "llm",
+      "cache_hit": false,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-43291-3",
       "nodes": [
@@ -16909,7 +16909,7 @@
         "pull_request:43291|pull_request:44815",
         "pull_request:43665|pull_request:44815"
       ],
-      "prepared_review_unit_hash": "2d1edca20f4033d07b7f4e9d3c2bf111b081c608ffc96229075859e56f3c4631",
+      "prepared_review_unit_hash": "cf40429a6a401b024316c5456f4429663061c9d1a3210e5e7a54e1a166b2f7fd",
       "component_budget": {
         "node_count": 3,
         "item_count": 3,
@@ -16931,30 +16931,30 @@
       "aggressively_trimmed": false,
       "split": false,
       "analyst_result": {
-        "summary": "These PRs are not duplicates: one is a Whisper tokenizer/test fix, one is a nonsensical CLIP/ViT test-only change, and one is a broader FP8/dequantization fix across loading and quantization code. They do not appear to address the same underlying bug or change.",
-        "confidence": 0.97,
+        "summary": "All three are pull requests, but they do not look like the same underlying change. PR 44815 is the only substantive bug-fix PR; 43291 is a Whisper tokenizer/test change, and 43665 is just placeholder test edits. No pair is close enough to merge as a duplicate.",
+        "confidence": 0.68,
         "canonical_issue_reason": null,
-        "canonical_pr_reason": "PR 44815 is the most substantive and concrete code fix in the set, with changes in model loading and FP8 quantization code plus related model/test updates. It is the best representative of a real bug-fix change, unlike the test-only or unclear PRs.",
+        "canonical_pr_reason": "PR 44815 is the strongest canonical candidate because it contains the only real implementation fix, spanning model loading and FP8 dequantization logic. The other two are test-focused and do not match its code path.",
         "best_issue_reason": null,
-        "best_pr_reason": "PR 44815 is the strongest standalone fix candidate because it targets a specific dequant/loading path and spans coherent implementation changes, making it the best representative of the cluster.",
+        "best_pr_reason": "PR 44815 is the best representative of the cluster: it has the clearest concrete bug fix, the broadest real code impact, and the most substantive diff. The others look like ancillary or unrelated test churn.",
         "soft_edge_verdicts": [
           {
             "left": "pull_request:43291",
             "right": "pull_request:43665",
             "accept": false,
-            "reason": "Both touch tests, but they fix different things: Whisper tokenizer decoding vs. unrelated CLIP/ViT test edits. No shared underlying bug or mergeable combined change is evident."
+            "reason": "Both touch tests, but 43291 also changes Whisper tokenizer behavior while 43665 is only placeholder assertions in unrelated model tests; they are not the same bug or change."
           },
           {
             "left": "pull_request:43291",
             "right": "pull_request:44815",
             "accept": false,
-            "reason": "Whisper tokenizer/timestamp decoding is unrelated to FP8 dequantization and model loading. Different subsystems and different code paths."
+            "reason": "43291 is a Whisper decoding/test fix, whereas 44815 is a dequantization/model-loading fix in a different subsystem; no shared concrete code-path problem."
           },
           {
             "left": "pull_request:43665",
             "right": "pull_request:44815",
             "accept": false,
-            "reason": "The CLIP/ViT test-only edits do not correspond to the FP8/core loading fix in 44815. They are not the same bug and would not plausibly be merged as one PR."
+            "reason": "43665 is clearly synthetic test churn, not a real implementation fix, so it is not the same underlying change as 44815."
           }
         ]
       },
@@ -16973,8 +16973,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "cache",
-      "cache_hit": true,
+      "source": "llm",
+      "cache_hit": false,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-43488-2",
       "nodes": [
@@ -16984,7 +16984,7 @@
       "soft_pairs": [
         "pull_request:43488|pull_request:43532"
       ],
-      "prepared_review_unit_hash": "91dcd6fd64a892094d356281aa2e99654c085319497a4940c9ba6d44236c869d",
+      "prepared_review_unit_hash": "8fe2f04d1e47126838a5f89e17d0e57c6ce649a512df17a8da4dcfb024707677",
       "component_budget": {
         "node_count": 2,
         "item_count": 2,
@@ -17006,18 +17006,18 @@
       "aggressively_trimmed": false,
       "split": false,
       "analyst_result": {
-        "summary": "Two open '[don't merge]' PRs, but they are not duplicates: one is a repo-bot/version-format test change, the other is a broader CI/workflow/diff-display change.",
-        "confidence": 0.97,
+        "summary": "These are not duplicates: PR #43488 is a synthetic repo-bot/checking change around setup/versioning and a Vit test, while PR #43532 rewires CI/workflow logic and touches several files. They only share a test filename, not the same underlying fix.",
+        "confidence": 0.96,
         "canonical_issue_reason": null,
-        "canonical_pr_reason": "No canonical PR: the PRs target different code paths and different purposes.",
+        "canonical_pr_reason": "PR #43532 is the better representative of the cluster because it is the more substantial, real change; PR #43488 is a placeholder-style bot check and does not match the same code path or change intent.",
         "best_issue_reason": null,
-        "best_pr_reason": "No clear best PR for a duplicate cluster; the shared Vit test file appears incidental rather than indicating the same fix.",
+        "best_pr_reason": "PR #43532 is the most suitable overall candidate: it has broader, coherent changes across CI and helper code, whereas #43488 is a small synthetic setup/test tweak.",
         "soft_edge_verdicts": [
           {
             "left": "pull_request:43488",
             "right": "pull_request:43532",
             "accept": false,
-            "reason": "Reject: 43488 changes setup/version/dependency test scaffolding, while 43532 rewires CI/comment workflows and testing utilities; they only overlap on a Vit test file and do not address the same underlying bug or change."
+            "reason": "Reject: the overlap is only a shared Vit test file. #43488 changes packaging/versioning and a test, while #43532 changes workflow triggers and related CI utilities; they do not fix the same concrete bug or belong in one PR."
           }
         ]
       },
@@ -17036,8 +17036,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "cache",
-      "cache_hit": true,
+      "source": "llm",
+      "cache_hit": false,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-43612-2",
       "nodes": [
@@ -17047,12 +17047,12 @@
       "soft_pairs": [
         "pull_request:43612|pull_request:43613"
       ],
-      "prepared_review_unit_hash": "bd93c4a494392cae4ab6176e63f753030a588929624482c823156a56b2e7414e",
+      "prepared_review_unit_hash": "7de38e7685f087524a82468bcff6beeae8336e22406c4b289d9dbbc74dbfac70",
       "component_budget": {
         "node_count": 2,
         "item_count": 2,
         "soft_pair_count": 1,
-        "serialized_chars": 6083,
+        "serialized_chars": 6082,
         "estimated_input_tokens": 1521,
         "estimated_eval_tokens": 3298
       },
@@ -17060,7 +17060,7 @@
         "node_count": 2,
         "item_count": 2,
         "soft_pair_count": 1,
-        "serialized_chars": 6083,
+        "serialized_chars": 6082,
         "estimated_input_tokens": 1521,
         "estimated_eval_tokens": 3298
       },
@@ -17069,18 +17069,18 @@
       "aggressively_trimmed": false,
       "split": false,
       "analyst_result": {
-        "summary": "These PRs are related by shared documentation/pipeline scaffolding, but they implement different features: Promptable Concept Segmentation vs Promptable Visual Segmentation. They should not be deduplicated.",
+        "summary": "The two PRs are structurally similar pipeline additions, but they implement different features: Promptable Concept Segmentation vs Promptable Visual Segmentation. They share documentation/metadata scaffolding, yet the model docs, pipeline modules, and tests target different code paths and use different underlying model families.",
         "confidence": 0.96,
         "canonical_issue_reason": null,
         "canonical_pr_reason": null,
         "best_issue_reason": null,
-        "best_pr_reason": "No single best PR for deduplication: they are separate pipeline additions with different code paths and model integrations.",
+        "best_pr_reason": null,
         "soft_edge_verdicts": [
           {
             "left": "pull_request:43612",
             "right": "pull_request:43613",
             "accept": false,
-            "reason": "Different underlying features and concrete implementations; shared docs/registry/test plumbing is not enough to treat them as the same change."
+            "reason": "Reject: these are not the same underlying bug or change. They add different pipelines with different task semantics and different model/test coverage, so they would not plausibly be merged into one PR."
           }
         ]
       },
@@ -17099,8 +17099,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "cache",
-      "cache_hit": true,
+      "source": "llm",
+      "cache_hit": false,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-43651-2",
       "nodes": [
@@ -17110,7 +17110,7 @@
       "soft_pairs": [
         "pull_request:43651|pull_request:43663"
       ],
-      "prepared_review_unit_hash": "6e0db7e77a75b6b0bba2280747342405b1dc3b46bebaae64f92cab1b6b2db6bb",
+      "prepared_review_unit_hash": "f062008cef8e56e673881cb4fb6961fc4641be5f95147c5ce3f00211e7b9ee6f",
       "component_budget": {
         "node_count": 2,
         "item_count": 2,
@@ -17132,18 +17132,18 @@
       "aggressively_trimmed": false,
       "split": false,
       "analyst_result": {
-        "summary": "The two PRs both touch `trainer.py` but implement unrelated Trainer extension points: one changes gradient-accumulation loss scaling, the other factorizes signature-column inspection for column filtering. They are not duplicates and should not be clustered together as the same change.",
+        "summary": "These two PRs both touch `trainer.py` and add extension hooks for custom trainers, but they solve different problems: gradient-accumulation loss scaling vs. column filtering based on the model signature. They should not be deduplicated.",
         "confidence": 0.98,
         "canonical_issue_reason": null,
-        "canonical_pr_reason": "No single canonical PR: the two PRs address different Trainer behaviors and do not share the same underlying bug or feature.",
+        "canonical_pr_reason": "No canonical PR: the changes target different trainer behaviors and are not the same underlying fix.",
         "best_issue_reason": null,
-        "best_pr_reason": "PR #43651 is the more broadly impactful Trainer behavior change and the clearer representative of the cluster, but it is still not a duplicate of #43663.",
+        "best_pr_reason": "Neither PR is a good global representative for a duplicate cluster because they are unrelated feature changes, not two versions of the same fix.",
         "soft_edge_verdicts": [
           {
             "left": "pull_request:43651",
             "right": "pull_request:43663",
             "accept": false,
-            "reason": "Different code paths and goals: #43651 changes loss scaling during training/gradient accumulation, while #43663 adds an overridable helper for model signature column filtering. Same file only; not the same underlying fix."
+            "reason": "Both modify `trainer.py`, but one adds `_loss_is_scaled_for_ga` for loss normalization and the other adds `_get_signature_columns` for dataset column filtering; they address different code paths and are not mergeable as a single duplicate PR."
           }
         ]
       },
@@ -17162,8 +17162,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "cache",
-      "cache_hit": true,
+      "source": "llm",
+      "cache_hit": false,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-43656-3",
       "nodes": [
@@ -17175,55 +17175,55 @@
         "pull_request:43656|pull_request:43842",
         "pull_request:43836|pull_request:43842"
       ],
-      "prepared_review_unit_hash": "18c490d394b5dd79ea75465ad7b7bfcdff5592c1fd6ea2f8b80b26a00fa70226",
+      "prepared_review_unit_hash": "f53ec09360f9bbba87f17d58c9446b6ebc6bb5503fa1c4f143337f326455a17f",
       "component_budget": {
         "node_count": 3,
         "item_count": 3,
         "soft_pair_count": 2,
-        "serialized_chars": 5991,
-        "estimated_input_tokens": 1498,
-        "estimated_eval_tokens": 3252
+        "serialized_chars": 5993,
+        "estimated_input_tokens": 1499,
+        "estimated_eval_tokens": 3254
       },
       "budget": {
         "node_count": 3,
         "item_count": 3,
         "soft_pair_count": 2,
-        "serialized_chars": 5991,
-        "estimated_input_tokens": 1498,
-        "estimated_eval_tokens": 3252
+        "serialized_chars": 5993,
+        "estimated_input_tokens": 1499,
+        "estimated_eval_tokens": 3254
       },
       "overflow_policy": "truncate_then_skip",
       "trimmed": false,
       "aggressively_trimmed": false,
       "split": false,
       "analyst_result": {
-        "summary": "All three PRs are overlapping fixes for `src/transformers/cli/serve.py` TypeAdapter/annotation `NameError` issues; one is a broader postponed-annotations cleanup and the others are narrower string-literal/import fixes.",
-        "confidence": 0.81,
+        "summary": "43842 is the cleanest representative of the serve.py TypeAdapter NameError fix. 43836 is a plausible duplicate/near-duplicate because it targets the same file and the same issue area, but 43656 is broader and less directly grounded as the same fix.",
+        "confidence": 0.78,
         "canonical_issue_reason": null,
-        "canonical_pr_reason": "PR 43842 is the cleanest, most targeted fix for the `TypeAdapter` NameError in `cli/serve.py`, with a minimal change focused on the failing runtime annotation path.",
+        "canonical_pr_reason": "PR 43842 is the most targeted patch for the TypeAdapter NameError in `src/transformers/cli/serve.py`, with a minimal diff and a direct issue link.",
         "best_issue_reason": null,
-        "best_pr_reason": "43842 is the best representative of the cluster because it directly addresses the concrete bug without the extra annotation cleanup/refactor seen in the other PRs.",
+        "best_pr_reason": "PR 43842 is the best cluster representative because it fixes the concrete TypeAdapter annotation problem with the least extra surface area.",
         "soft_edge_verdicts": [
           {
             "left": "pull_request:43656",
             "right": "pull_request:43842",
-            "accept": true,
-            "reason": "Both address the same runtime `TypeAdapter` annotation NameError in `src/transformers/cli/serve.py`; they differ only in implementation approach (future annotations vs quoted/type-only import handling)."
+            "accept": false,
+            "reason": "Both touch `serve.py`, but 43656\u2019s `__future__`/annotation rewrite is broader and its explicit issue target differs, so the evidence is not strong enough to call it the same concrete fix as 43842."
           },
           {
             "left": "pull_request:43836",
             "right": "pull_request:43842",
             "accept": true,
-            "reason": "Both appear to fix the same underlying `TypeAdapter`/annotation NameError in the CLI serve code path and could plausibly be merged into one PR."
+            "reason": "These are directly aligned on the same `serve.py` TypeAdapter NameError path, and 43836\u2019s broader annotation cleanup still appears to be addressing the same runtime typing issue."
           }
         ]
       },
       "evaluator_result": {
-        "accept": true,
-        "feedback": "The summary is grounded in the packet: all three PRs touch `src/transformers/cli/serve.py` and are about annotation/`TypeAdapter` NameError fixes. The pairwise soft-edge calls are conservative enough because the changes are mergeable into a single fix for the same bug family, with 43656 being the broader postponed-annotations variant and 43836/43842 being narrower string-literal/import variants."
+        "accept": false,
+        "feedback": "The summary overstates what the packet supports. All three PRs touch the same file, but the evidence does not establish that they are all the same TypeAdapter NameError fix: 43656\u2019s preview shows a broader `__future__`/annotation rewrite and it has a different explicit issue target, so calling it a close variant of 43842 is too strong. The 43836\u219443842 pair is better supported, but the analyst still presents the cluster as one uniform bugfix without enough grounding. Please narrow the claims to what is directly evidenced and be more conservative on the soft-edge pair for 43656."
       },
       "evaluator_used": true,
-      "retried": false,
+      "retried": true,
       "accepted_nontrivial_soft_edge": true,
       "error_kind": null,
       "error_message": null
@@ -17302,8 +17302,8 @@
       "review_unit_count": 2,
       "status": "reviewed",
       "reason": null,
-      "source": "cache",
-      "cache_hit": true,
+      "source": "llm",
+      "cache_hit": false,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-43996-28",
       "nodes": [
@@ -17409,20 +17409,20 @@
         "pull_request:44044|pull_request:44129",
         "pull_request:44044|pull_request:44154"
       ],
-      "prepared_review_unit_hash": "5537f499b6d97d82d54f2671563ab9b7a753a6173e9490c103b1a5d0ad596b68",
+      "prepared_review_unit_hash": "394c9d1e321da9e348237be25e57d6ad2fb10665530a1ce559509a9fb7313438",
       "component_budget": {
         "node_count": 28,
         "item_count": 28,
         "soft_pair_count": 78,
-        "serialized_chars": 84898,
-        "estimated_input_tokens": 21225,
-        "estimated_eval_tokens": 42706
+        "serialized_chars": 84904,
+        "estimated_input_tokens": 21226,
+        "estimated_eval_tokens": 42708
       },
       "budget": {
         "node_count": 27,
         "item_count": 27,
         "soft_pair_count": 72,
-        "serialized_chars": 75098,
+        "serialized_chars": 75099,
         "estimated_input_tokens": 18775,
         "estimated_eval_tokens": 37806
       },
@@ -17431,412 +17431,424 @@
       "aggressively_trimmed": false,
       "split": true,
       "analyst_result": {
-        "summary": "Mostly parallel, model-specific output-tracing refactors. The only convincing duplicate family in the packet is the GPT-J/CodeGen line around 44722; the GPT-Neo and other shared-issue edges look like similarity neighbors, not mergeable duplicates.",
-        "confidence": 0.67,
-        "canonical_issue_reason": null,
-        "canonical_pr_reason": "PR 44722 is the best representative of the GPT-J/CodeGen subgroup: it is the later, broader revision in the same files and the clearest true-duplicate candidate in the set.",
-        "best_issue_reason": null,
-        "best_pr_reason": "PR 44722 is the strongest single representative because it covers the same GPT-J/CodeGen code paths as the closest neighbors and looks like the most complete later revision.",
+        "summary": "This cluster is mainly a shared output-tracing refactor stream across many models. The common umbrella is issue #43979, but only a few pairs are true duplicates at the PR level: same model/file and same concrete output-capture change.",
+        "confidence": 0.89,
+        "canonical_issue_reason": "Shared tracker #43979 is the umbrella issue for the standardized output-tracing migration across these PRs.",
+        "canonical_pr_reason": "PR 44066 is the best representative merge anchor: it explicitly targets #43979, makes a concrete GPT-J/CodeGen output-tracing refactor, and has the clearest direct duplicate overlap with 44722 and 44085.",
+        "best_issue_reason": "Issue #43979 is the global umbrella for this refactor line and is the natural canonical issue for deduping the related PRs.",
+        "best_pr_reason": "PR 44066 is the strongest single PR representative because it is explicit, concrete, and closest to the duplicated GPT-J follow-ups.",
         "soft_edge_verdicts": [
           {
             "left": "pull_request:43996",
             "right": "pull_request:44085",
             "accept": false,
-            "reason": "Different model families; the shared issue theme is too broad to make this a duplicate."
+            "reason": "Different models/files; both are output-tracing refactors, but not the same code path."
           },
           {
             "left": "pull_request:43996",
             "right": "pull_request:44044",
             "accept": false,
-            "reason": "Different model code paths; same tracking issue theme is not enough."
+            "reason": "Different model families and different concrete implementations."
           },
           {
             "left": "pull_request:44066",
             "right": "pull_request:44085",
-            "accept": false,
-            "reason": "44066 spans CodeGen + GPT-J while 44085 is GPT-J-only; similar, but not clearly the same concrete change."
+            "accept": true,
+            "reason": "Same GPT-J output-tracing code path in the same file; 44066 is a broader version of the same fix."
           },
           {
             "left": "pull_request:44007",
             "right": "pull_request:44072",
             "accept": false,
-            "reason": "Different backbones/model implementations; parallel refactors, not duplicates."
+            "reason": "Different backbones and different files; only the refactor theme is shared."
           },
           {
             "left": "pull_request:44072",
             "right": "pull_request:44722",
             "accept": false,
-            "reason": "EfficientNet is unrelated to GPT-J/CodeGen output tracing."
+            "reason": "EfficientNet versus GPT-J/CodeGen; unrelated concrete changes."
           },
           {
             "left": "pull_request:44066",
             "right": "pull_request:44072",
             "accept": false,
-            "reason": "Different model families; same refactor theme only."
+            "reason": "Different models and different code paths."
           },
           {
             "left": "pull_request:44013",
             "right": "pull_request:44044",
             "accept": false,
-            "reason": "MobileNetV2 and DeBERTa v2 are unrelated code paths."
+            "reason": "MobileNetV2 and DeBERTa-v2 are unrelated implementations."
           },
           {
             "left": "pull_request:44066",
             "right": "pull_request:44086",
             "accept": false,
-            "reason": "GPT-J/CodeGen refactor vs MGP-STR refactor; different underlying change."
+            "reason": "Different model code paths; same thematic refactor only."
           },
           {
             "left": "pull_request:44066",
             "right": "pull_request:44071",
             "accept": false,
-            "reason": "Same standardized-output theme, but MPT and GPT-J/CodeGen are different implementations."
+            "reason": "GPT-J/CodeGen versus MPT; not the same bug or change."
           },
           {
             "left": "pull_request:44018",
             "right": "pull_request:44068",
-            "accept": false,
-            "reason": "Same GPT-Neo area, but the packet only shows similarity and not a clearly mergeable identical edit."
+            "accept": true,
+            "reason": "Both are GPT-Neo output-tracing refactors on the same concrete code path."
           },
           {
             "left": "pull_request:44066",
             "right": "pull_request:44068",
             "accept": false,
-            "reason": "GPT-J/CodeGen vs GPT-Neo are separate model-specific refactors."
+            "reason": "GPT-J/CodeGen and GPT-Neo are separate implementations."
           },
           {
             "left": "pull_request:44018",
             "right": "pull_request:44066",
             "accept": false,
-            "reason": "GPT-Neo and GPT-J/CodeGen are different code paths."
+            "reason": "GPT-Neo versus GPT-J/CodeGen; same pattern, different code path."
           },
           {
             "left": "pull_request:44019",
             "right": "pull_request:44722",
             "accept": false,
-            "reason": "ResNet refactor is unrelated to GPT-J/CodeGen output tracing."
+            "reason": "ResNet and GPT-J/CodeGen are unrelated model implementations."
           },
           {
             "left": "pull_request:44086",
             "right": "pull_request:44722",
             "accept": false,
-            "reason": "MGP-STR and GPT-J/CodeGen are unrelated model changes."
+            "reason": "MGP-STR and GPT-J/CodeGen are different concrete changes."
           },
           {
             "left": "pull_request:44019",
             "right": "pull_request:44085",
             "accept": false,
-            "reason": "ResNet and GPT-J are separate model-specific changes."
+            "reason": "ResNet versus GPT-J; only the refactor style matches."
           },
           {
             "left": "pull_request:44018",
             "right": "pull_request:44019",
             "accept": false,
-            "reason": "GPT-Neo and ResNet are unrelated implementations."
+            "reason": "GPT-Neo and ResNet are different models with different output paths."
           },
           {
             "left": "pull_request:44019",
             "right": "pull_request:44071",
             "accept": false,
-            "reason": "ResNet and MPT are different model code paths."
+            "reason": "Different model code paths."
           },
           {
             "left": "pull_request:44068",
             "right": "pull_request:44116",
             "accept": false,
-            "reason": "GPT-Neo and Flaubert are separate model refactors."
+            "reason": "GPT-Neo versus Flaubert; unrelated implementations."
           },
           {
             "left": "pull_request:44068",
             "right": "pull_request:44722",
             "accept": false,
-            "reason": "GPT-Neo and GPT-J/CodeGen are distinct model-specific changes."
+            "reason": "GPT-Neo and GPT-J/CodeGen are different concrete refactors."
           },
           {
             "left": "pull_request:44018",
             "right": "pull_request:44116",
             "accept": false,
-            "reason": "GPT-Neo and Flaubert are unrelated refactors."
+            "reason": "GPT-Neo and Flaubert are unrelated code paths."
           },
           {
             "left": "pull_request:44071",
-            "right": "pull_request:44722",
-            "accept": false,
-            "reason": "MPT and GPT-J/CodeGen are different model code paths."
-          },
-          {
-            "left": "pull_request:44018",
             "right": "pull_request:44086",
             "accept": false,
-            "reason": "GPT-Neo and MGP-STR are unrelated implementations."
-          },
-          {
-            "left": "pull_request:44018",
-            "right": "pull_request:44085",
-            "accept": false,
-            "reason": "GPT-Neo and GPT-J are separate model-specific refactors."
+            "reason": "MPT and MGP-STR are different model implementations."
           },
           {
-            "left": "pull_request:44068",
-            "right": "pull_request:44086",
+            "left": "pull_request:44071",
+            "right": "pull_request:44722",
             "accept": false,
-            "reason": "GPT-Neo and MGP-STR are different code paths."
+            "reason": "MPT versus GPT-J/CodeGen; not mergeable as one PR."
           },
           {
             "left": "pull_request:44059",
             "right": "pull_request:44086",
             "accept": false,
-            "reason": "GPT-2 and MGP-STR are unrelated model changes."
+            "reason": "GPT2 and MGP-STR are different concrete changes."
           },
           {
             "left": "pull_request:44085",
             "right": "pull_request:44086",
             "accept": false,
-            "reason": "GPT-J and MGP-STR are different implementations."
+            "reason": "GPT-J and MGP-STR are unrelated code paths."
           },
           {
             "left": "pull_request:44059",
             "right": "pull_request:44722",
             "accept": false,
-            "reason": "GPT-2 refactor is unrelated to GPT-J/CodeGen."
+            "reason": "GPT2 versus GPT-J/CodeGen; same theme, different implementation."
           },
           {
             "left": "pull_request:44098",
             "right": "pull_request:44101",
             "accept": false,
-            "reason": "ViLT and XLM/Flaubert are different model families."
+            "reason": "ViLT and Flaubert/XLM are different model families."
           },
           {
             "left": "pull_request:44018",
             "right": "pull_request:44071",
             "accept": false,
-            "reason": "GPT-Neo and MPT are separate refactors."
+            "reason": "GPT-Neo and MPT are unrelated concrete refactors."
           },
           {
             "left": "pull_request:44017",
             "right": "pull_request:44161",
             "accept": false,
-            "reason": "SegFormer and LongT5 are unrelated model-specific changes."
+            "reason": "SegFormer and LongT5 are unrelated model implementations."
           },
           {
             "left": "pull_request:44017",
             "right": "pull_request:44722",
             "accept": false,
-            "reason": "SegFormer is unrelated to GPT-J/CodeGen output tracing."
+            "reason": "SegFormer versus GPT-J/CodeGen; not the same bug or change."
           },
           {
             "left": "pull_request:44017",
             "right": "pull_request:44073",
             "accept": false,
-            "reason": "SegFormer and VisualBert are different implementations."
+            "reason": "SegFormer and VisualBERT are different code paths."
           },
           {
             "left": "pull_request:44068",
             "right": "pull_request:44071",
             "accept": false,
-            "reason": "GPT-Neo and MPT are different model code paths."
+            "reason": "GPT-Neo and MPT are different models."
           },
           {
             "left": "pull_request:44071",
             "right": "pull_request:44086",
             "accept": false,
-            "reason": "MPT and MGP-STR are unrelated refactors."
-          },
-          {
-            "left": "pull_request:43997",
-            "right": "pull_request:44073",
-            "accept": false,
-            "reason": "RegNet and VisualBert are different model-specific changes."
-          },
-          {
-            "left": "pull_request:44017",
-            "right": "pull_request:44071",
-            "accept": false,
-            "reason": "SegFormer and MPT are unrelated model refactors."
-          },
-          {
-            "left": "pull_request:44073",
-            "right": "pull_request:44101",
-            "accept": false,
-            "reason": "VisualBert and Flaubert are different code paths."
+            "reason": "MPT and MGP-STR are unrelated implementations."
           },
           {
             "left": "pull_request:44059",
             "right": "pull_request:44071",
             "accept": false,
-            "reason": "GPT-2 and MPT are separate refactors."
+            "reason": "GPT2 and MPT are different code paths."
           },
           {
             "left": "pull_request:44073",
-            "right": "pull_request:44161",
+            "right": "pull_request:44101",
             "accept": false,
-            "reason": "VisualBert and LongT5 are unrelated changes."
+            "reason": "VisualBERT and Flaubert/XLM are different models."
           },
           {
             "left": "pull_request:44056",
             "right": "pull_request:44101",
             "accept": false,
-            "reason": "MPNet and Flaubert are separate model-specific refactors."
+            "reason": "MPNet and Flaubert/XLM are different implementations."
           },
           {
             "left": "pull_request:44059",
             "right": "pull_request:44068",
             "accept": false,
-            "reason": "GPT-2 and GPT-Neo are different model code paths."
+            "reason": "GPT2 and GPT-Neo are separate model code paths."
           },
           {
             "left": "pull_request:44018",
             "right": "pull_request:44059",
             "accept": false,
-            "reason": "GPT-Neo and GPT-2 are distinct refactors."
+            "reason": "GPT-Neo versus GPT2; same refactor theme only."
           },
           {
             "left": "pull_request:44101",
             "right": "pull_request:44161",
             "accept": false,
-            "reason": "Flaubert and LongT5 are unrelated model changes."
+            "reason": "Flaubert/XLM and LongT5 are unrelated changes."
           },
           {
             "left": "pull_request:43997",
             "right": "pull_request:44017",
             "accept": false,
-            "reason": "RegNet and SegFormer are different model-specific refactors."
+            "reason": "RegNet and SegFormer are different model families."
           },
           {
             "left": "pull_request:43997",
             "right": "pull_request:44056",
             "accept": false,
-            "reason": "RegNet and MPNet are separate code paths."
+            "reason": "RegNet and MPNet are unrelated."
           },
           {
             "left": "pull_request:44071",
             "right": "pull_request:44076",
             "accept": false,
-            "reason": "MPT and ImageGPT are unrelated model implementations."
+            "reason": "MPT and ImageGPT are different concrete implementations."
           },
           {
             "left": "pull_request:44074",
             "right": "pull_request:44101",
             "accept": false,
-            "reason": "TextNet and Flaubert are different model refactors."
+            "reason": "TextNet and Flaubert/XLM are unrelated model code paths."
           },
           {
             "left": "pull_request:44056",
             "right": "pull_request:44161",
             "accept": false,
-            "reason": "MPNet and LongT5 are unrelated changes."
+            "reason": "MPNet and LongT5 are different implementations."
           },
           {
             "left": "pull_request:43997",
             "right": "pull_request:44018",
             "accept": false,
-            "reason": "RegNet and GPT-Neo are unrelated model-specific refactors."
+            "reason": "RegNet and GPT-Neo are unrelated model code paths."
           },
           {
             "left": "pull_request:43997",
             "right": "pull_request:44161",
             "accept": false,
-            "reason": "RegNet and LongT5 are different model code paths."
+            "reason": "RegNet and LongT5 are different models."
           },
           {
             "left": "pull_request:44074",
             "right": "pull_request:44161",
             "accept": false,
-            "reason": "TextNet and LongT5 are unrelated refactors."
+            "reason": "TextNet and LongT5 are unrelated implementations."
           },
           {
             "left": "pull_request:44018",
             "right": "pull_request:44161",
             "accept": false,
-            "reason": "GPT-Neo and LongT5 are separate model changes."
+            "reason": "GPT-Neo and LongT5 are different model code paths."
           },
           {
             "left": "pull_request:43997",
             "right": "pull_request:44071",
             "accept": false,
-            "reason": "RegNet and MPT are different implementations."
+            "reason": "RegNet and MPT are unrelated."
           },
           {
             "left": "pull_request:44010",
             "right": "pull_request:44071",
             "accept": false,
-            "reason": "SqueezeBert and MPT are unrelated model refactors."
+            "reason": "SqueezeBERT and MPT are different models."
           },
           {
             "left": "pull_request:44010",
             "right": "pull_request:44073",
             "accept": false,
-            "reason": "SqueezeBert and VisualBert are different code paths."
+            "reason": "SqueezeBERT and VisualBERT are different concrete changes."
           },
           {
             "left": "pull_request:44018",
             "right": "pull_request:44129",
             "accept": false,
-            "reason": "GPT-Neo and SpeechT5 are unrelated model-specific changes."
+            "reason": "GPT-Neo and SpeechT5 are unrelated model implementations."
+          },
+          {
+            "left": "pull_request:44074",
+            "right": "pull_request:44161",
+            "accept": false,
+            "reason": "TextNet and LongT5 are different code paths."
+          },
+          {
+            "left": "pull_request:43997",
+            "right": "pull_request:44074",
+            "accept": false,
+            "reason": "RegNet and TextNet are unrelated models."
+          },
+          {
+            "left": "pull_request:43997",
+            "right": "pull_request:44129",
+            "accept": false,
+            "reason": "RegNet and SpeechT5 are different implementations."
+          },
+          {
+            "left": "pull_request:44071",
+            "right": "pull_request:44129",
+            "accept": false,
+            "reason": "MPT and SpeechT5 are unrelated."
+          },
+          {
+            "left": "pull_request:44010",
+            "right": "pull_request:44018",
+            "accept": false,
+            "reason": "SqueezeBERT and GPT-Neo are different model code paths."
+          },
+          {
+            "left": "pull_request:43996",
+            "right": "pull_request:44161",
+            "accept": false,
+            "reason": "FNet/CVT and LongT5 are unrelated."
+          },
+          {
+            "left": "pull_request:43996",
+            "right": "pull_request:44017",
+            "accept": false,
+            "reason": "FNet/CVT and SegFormer are different model families."
           },
           {
             "left": "pull_request:44010",
             "right": "pull_request:44044",
             "accept": false,
-            "reason": "SqueezeBert and DeBERTa v2 are separate model refactors."
+            "reason": "SqueezeBERT and DeBERTa-v2 are unrelated."
           },
           {
             "left": "pull_request:44044",
             "right": "pull_request:44074",
             "accept": false,
-            "reason": "DeBERTa v2 and TextNet are unrelated changes."
+            "reason": "DeBERTa-v2 and TextNet are different concrete changes."
           },
           {
             "left": "pull_request:44066",
             "right": "pull_request:44722",
             "accept": true,
-            "reason": "Same GPT-J/CodeGen output-tracing refactor in the same files; 44722 looks like the later, more complete revision."
+            "reason": "Same GPT-J/CodeGen output-tracing refactor; 44722 is a near-duplicate follow-up."
           },
           {
             "left": "pull_request:44085",
             "right": "pull_request:44722",
             "accept": true,
-            "reason": "Same GPT-J output-tracing change in modeling_gptj.py; 44722 appears to be the later revision."
+            "reason": "Both target the same GPT-J output-tracing code path and could be merged into one PR."
           },
           {
             "left": "pull_request:44085",
             "right": "pull_request:44129",
             "accept": false,
-            "reason": "GPT-J refactor and SpeechT5 refactor are unrelated."
+            "reason": "GPT-J and SpeechT5 are different implementations."
           },
           {
             "left": "pull_request:43996",
             "right": "pull_request:44007",
             "accept": false,
-            "reason": "CVT/FNet refactor and RegNet/ResNet refactor are different changes."
+            "reason": "FNet/CVT and RegNet/ResNet/RT-DETR are unrelated."
           },
           {
             "left": "pull_request:43996",
             "right": "pull_request:44129",
             "accept": false,
-            "reason": "CVT/FNet and SpeechT5 are unrelated model refactors."
+            "reason": "FNet/CVT and SpeechT5 are different model code paths."
           },
           {
             "left": "pull_request:44044",
             "right": "pull_request:44129",
             "accept": false,
-            "reason": "DeBERTa v2 and SpeechT5 are different code paths."
+            "reason": "DeBERTa-v2 and SpeechT5 are unrelated implementations."
           },
           {
             "left": "pull_request:44044",
             "right": "pull_request:44154",
             "accept": false,
-            "reason": "DeBERTa v2 and Vits are unrelated model changes."
+            "reason": "DeBERTa-v2 and ViTS are different concrete changes."
           }
         ]
       },
       "evaluator_result": {
-        "accept": false,
-        "feedback": "The cluster read is mostly grounded, but two soft-edge calls are too assertive for the evidence shown. In particular, 44018\u219444068 and 44066\u219444085 are treated as clear duplicates, yet the packet only supports them as similarity/issue-target neighbors, not clearly mergeable identical changes; 44066 also spans CodeGen + GPT-J while 44085 is GPT-J-only. The summary also overstates that the remaining PRs all share the same tracking issue, when several have no explicit issue target in the packet. A more conservative writeup is needed."
+        "accept": true,
+        "feedback": "The cluster summary is grounded in the packet: #43979 is the shared issue target across the refactor stream, and the representative PR choice is supported by the GPT-J/CodeGen overlap. The soft-edge verdicts are conservative overall, and the true-accept pairs are limited to same-model/code-path cases that do look mergeable into a single bugfix PR."
       },
       "evaluator_used": true,
-      "retried": true,
-      "accepted_nontrivial_soft_edge": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": true,
       "error_kind": null,
       "error_message": null
     },
@@ -17848,8 +17860,8 @@
       "review_unit_count": 2,
       "status": "reviewed",
       "reason": null,
-      "source": "cache",
-      "cache_hit": true,
+      "source": "llm",
+      "cache_hit": false,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-43996-28",
       "nodes": [
@@ -17870,70 +17882,70 @@
         "pull_request:43997|pull_request:44010",
         "pull_request:44010|pull_request:44129"
       ],
-      "prepared_review_unit_hash": "508780701f326b01f6789058b66f4e1701fc55188c4a807be582ea20f6f755f8",
+      "prepared_review_unit_hash": "445c8ef2736f689f9e0062f64ab50a2fc6932c0e9fe647dae3963e1c0e91db7f",
       "component_budget": {
         "node_count": 28,
         "item_count": 28,
         "soft_pair_count": 78,
-        "serialized_chars": 84898,
-        "estimated_input_tokens": 21225,
-        "estimated_eval_tokens": 42706
+        "serialized_chars": 84904,
+        "estimated_input_tokens": 21226,
+        "estimated_eval_tokens": 42708
       },
       "budget": {
         "node_count": 8,
         "item_count": 8,
         "soft_pair_count": 6,
-        "serialized_chars": 17121,
-        "estimated_input_tokens": 4281,
-        "estimated_eval_tokens": 8818
+        "serialized_chars": 17126,
+        "estimated_input_tokens": 4282,
+        "estimated_eval_tokens": 8820
       },
       "overflow_policy": "truncate_then_skip",
       "trimmed": false,
       "aggressively_trimmed": false,
       "split": true,
       "analyst_result": {
-        "summary": "These are model-specific PRs for the same broad refactor theme\u2014standardized output tracing/capture\u2014but they do not appear to be duplicates of one another.",
-        "confidence": 0.93,
+        "summary": "These PRs are all separate per-model migrations to standardized output-capturing/tracing patterns. They share a common refactor theme, but they do not appear to be the same concrete fix or mergeable into one PR.",
+        "confidence": 0.94,
         "canonical_issue_reason": null,
-        "canonical_pr_reason": "PR 44129 is the clearest representative: it has an explicit issue link, a focused SpeechT5 output-tracing refactor, and concrete code-path changes rather than just naming the pattern.",
+        "canonical_pr_reason": "pull_request:44129 is the strongest representative: it has an explicit issue target, a concrete model-specific refactor, and the most clearly scoped change among the set.",
         "best_issue_reason": null,
-        "best_pr_reason": "PR 44129 is the strongest standalone example of the cluster because it is well-scoped, explicitly tracked, and most directly describes the standardized output-capture change.",
+        "best_pr_reason": "pull_request:44129 is the best overall PR because it is well-scoped, explicitly linked to an issue, and clearly documents a concrete output-tracing migration for one code path.",
         "soft_edge_verdicts": [
           {
             "left": "pull_request:44024",
             "right": "pull_request:44076",
             "accept": false,
-            "reason": "Both are output-tracing refactors, but they modify different model code paths (FocalNet vs ImageGPT) and are separate model-specific migrations, not one underlying bug or change."
+            "reason": "Both are standardized output-tracing refactors, but for different models (FocalNet vs ImageGPT) and different code paths; not the same bug/change."
           },
           {
             "left": "pull_request:44073",
             "right": "pull_request:44074",
             "accept": false,
-            "reason": "VisualBert and TextNet are different models with independent refactors; sharing the capture_outputs pattern is too broad to call them duplicates."
+            "reason": "VisualBert and TextNet are unrelated model implementations; these are parallel refactors, not one shared fix."
           },
           {
             "left": "pull_request:44056",
             "right": "pull_request:44073",
             "accept": false,
-            "reason": "MPNet and VisualBert touch different architectures and layer stacks, so these are parallel refactors rather than the same concrete fix."
+            "reason": "MPNet and VisualBert have different forward paths and output-recording setup; same refactor theme only, not duplicate work."
           },
           {
             "left": "pull_request:44056",
             "right": "pull_request:44074",
             "accept": false,
-            "reason": "These target different model implementations and do not appear to fix the same code-path problem."
+            "reason": "MPNet and TextNet are separate models with distinct output-capture changes; not the same underlying code-path problem."
           },
           {
             "left": "pull_request:43997",
             "right": "pull_request:44010",
             "accept": false,
-            "reason": "RegNet and SqueezeBert are unrelated model-specific output-tracing updates; same theme, different changes."
+            "reason": "RegNet and SqueezeBert are unrelated model modules; these PRs do not fix the same concrete issue."
           },
           {
             "left": "pull_request:44010",
             "right": "pull_request:44129",
             "accept": false,
-            "reason": "SqueezeBert and SpeechT5 are separate model refactors, and 44129 is tied to its own issue target, so they should not be merged as duplicates."
+            "reason": "SqueezeBert and SpeechT5 both touch output-tracing, but they affect different model internals and are not plausibly one merged PR."
           }
         ]
       },
@@ -17952,8 +17964,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "cache",
-      "cache_hit": true,
+      "source": "llm",
+      "cache_hit": false,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-43998-13",
       "nodes": [
@@ -18013,256 +18025,274 @@
         "pull_request:44028|pull_request:44030",
         "pull_request:44029|pull_request:44030"
       ],
-      "prepared_review_unit_hash": "caf725ee08d90c5d2e85b14bb0a9d6912127158cfe8cec3a8035a8d9a545c960",
+      "prepared_review_unit_hash": "28c4cef5238fb6480721688abb6d4913ff90fd6e0deaaabfea8559a681ab8e12",
       "component_budget": {
         "node_count": 13,
         "item_count": 13,
         "soft_pair_count": 40,
-        "serialized_chars": 40169,
-        "estimated_input_tokens": 10043,
-        "estimated_eval_tokens": 20342
+        "serialized_chars": 40179,
+        "estimated_input_tokens": 10045,
+        "estimated_eval_tokens": 20346
       },
       "budget": {
         "node_count": 13,
         "item_count": 13,
         "soft_pair_count": 40,
-        "serialized_chars": 40169,
-        "estimated_input_tokens": 10043,
-        "estimated_eval_tokens": 20342
+        "serialized_chars": 40179,
+        "estimated_input_tokens": 10045,
+        "estimated_eval_tokens": 20346
       },
       "overflow_policy": "truncate_then_skip",
       "trimmed": false,
       "aggressively_trimmed": false,
       "split": false,
       "analyst_result": {
-        "summary": "These are all model-specific \"refactor output tracing\" PRs that share a template but target different codepaths. I found no true duplicates among the soft-edge candidates, so all proposed matches are rejected.",
-        "confidence": 0.97,
+        "summary": "These PRs share a naming pattern and a broad \u201coutput tracing\u201d theme, but they are model-specific refactors in different code paths. I do not see true duplicate PRs here.",
+        "confidence": 0.95,
         "canonical_issue_reason": null,
-        "canonical_pr_reason": "PR 44003 is the strongest representative of the cluster: it has the broadest concrete output-capturing change, spans related files, and is more technically substantial than the single-model refactors.",
+        "canonical_pr_reason": "No single true duplicate PR stands out; 44025 is the best representative anchor because it\u2019s a clear output-tracing refactor across two related model files, but it is still not the same change as the others.",
         "best_issue_reason": null,
-        "best_pr_reason": "PR 44003 is the best overall exemplar for this cluster because it shows the clearest, most central instance of the recurring output-tracing refactor pattern with explicit capture plumbing.",
+        "best_pr_reason": "44025 is the strongest overall representative of the cluster\u2019s theme, with a broader two-file refactor footprint and central similarity to several others, even though it should not be treated as a duplicate of them.",
         "soft_edge_verdicts": [
           {
             "left": "pull_request:43998",
             "right": "pull_request:44002",
             "accept": false,
-            "reason": "Different model families (timm_backbone vs upernet); same refactor style, but not the same underlying change or mergeable PR."
+            "reason": "Different models and code paths; same refactor theme only, not the same bug/change."
           },
           {
             "left": "pull_request:43998",
             "right": "pull_request:44000",
             "accept": false,
-            "reason": "Different codepaths and architectures; this is only a shared output-tracing refactor pattern."
+            "reason": "Separate model implementations; similarity is stylistic, not a shared concrete fix."
           },
           {
             "left": "pull_request:43998",
             "right": "pull_request:44001",
             "accept": false,
-            "reason": "timm_backbone and univnet are unrelated model implementations; not the same bug/change."
+            "reason": "Different architecture and file; not the same underlying change."
           },
           {
             "left": "pull_request:43998",
             "right": "pull_request:44003",
             "accept": false,
-            "reason": "Different model families (timm_backbone vs mamba/falcon_mamba); similar decorator/capture refactor, but not one duplicate change."
-          },
-          {
-            "left": "pull_request:43998",
-            "right": "pull_request:44004",
-            "accept": false,
-            "reason": "Different architectures and codepaths; only the refactor pattern overlaps."
+            "reason": "Unrelated model code paths; only the refactor pattern matches."
           },
           {
             "left": "pull_request:43998",
             "right": "pull_request:44029",
             "accept": false,
-            "reason": "Different model families (timm_backbone vs rwkv); same broad theme, not the same concrete fix."
+            "reason": "Different model family; no evidence of one merged fix."
           },
           {
             "left": "pull_request:43999",
             "right": "pull_request:44001",
             "accept": false,
-            "reason": "Different models (mobilenet_v1 vs univnet); shared output-tracing cleanup only."
+            "reason": "Different models; shared output-tracing cleanup is too broad to count as duplicate."
           },
           {
             "left": "pull_request:43999",
             "right": "pull_request:44002",
             "accept": false,
-            "reason": "Different model families (mobilenet_v1 vs upernet); not the same underlying change."
+            "reason": "Separate model code paths with no concrete overlap beyond refactor style."
           },
           {
             "left": "pull_request:43999",
             "right": "pull_request:44025",
             "accept": false,
-            "reason": "mobilenet_v1 and depth_anything are distinct model implementations; this is not a duplicate PR."
+            "reason": "Different model families; not the same code-path problem."
           },
           {
             "left": "pull_request:43999",
             "right": "pull_request:44026",
             "accept": false,
-            "reason": "Different wrappers/models; only the refactor template is shared."
+            "reason": "Distinct encoder-decoder vs backbone model changes; not duplicate work."
           },
           {
             "left": "pull_request:43999",
             "right": "pull_request:44027",
             "accept": false,
-            "reason": "mobilenet_v1 vs speech_encoder_decoder are unrelated codepaths; reject."
+            "reason": "Same generic tracing refactor, but different subsystem and behavior."
           },
           {
             "left": "pull_request:44000",
             "right": "pull_request:44001",
             "accept": false,
-            "reason": "Different models (vision_text_dual_encoder vs univnet); similar output-handling refactor, but not one change."
+            "reason": "Different models; too generic to be a duplicate."
           },
           {
             "left": "pull_request:44000",
             "right": "pull_request:44002",
             "accept": false,
-            "reason": "Different model families and files; no shared concrete bug or fix."
+            "reason": "Separate code paths and model classes; only the theme overlaps."
           },
           {
             "left": "pull_request:44000",
             "right": "pull_request:44003",
             "accept": false,
-            "reason": "Different model stacks (vision_text_dual_encoder vs mamba/falcon_mamba); not mergeable as one PR."
+            "reason": "No shared concrete bug or implementation target."
           },
           {
             "left": "pull_request:44000",
             "right": "pull_request:44025",
             "accept": false,
-            "reason": "Different model families; same tracing refactor pattern only."
+            "reason": "Different model families; same refactor motif is not enough."
           },
           {
             "left": "pull_request:44001",
             "right": "pull_request:44002",
             "accept": false,
-            "reason": "univnet and upernet are unrelated; this is only superficial similarity."
+            "reason": "Different architectures; not plausibly one PR."
           },
           {
             "left": "pull_request:44001",
-            "right": "pull_request:44025",
+            "right": "pull_request:44003",
             "accept": false,
-            "reason": "Different model codepaths (univnet vs depth_anything/prompt_depth_anything); not a duplicate."
+            "reason": "No shared failing code-path, just similar cleanup."
           },
           {
             "left": "pull_request:44001",
-            "right": "pull_request:44004",
+            "right": "pull_request:44025",
             "accept": false,
-            "reason": "Different architectures; shared refactor motif is not enough to merge."
+            "reason": "Distinct model-specific output handling; not the same change."
+          },
+          {
+            "left": "pull_request:44002",
+            "right": "pull_request:44029",
+            "accept": false,
+            "reason": "Different model implementations and output semantics; not duplicate."
           },
           {
             "left": "pull_request:44002",
             "right": "pull_request:44003",
             "accept": false,
-            "reason": "upernet and mamba/falcon_mamba are different model families; reject."
+            "reason": "Same refactor category, but different concrete fixes."
           },
           {
             "left": "pull_request:44003",
             "right": "pull_request:44028",
             "accept": false,
-            "reason": "Different model families (mamba/falcon_mamba vs superpoint); same output-tracing theme, but not the same concrete change."
+            "reason": "Mamba vs SuperPoint are unrelated code paths; no duplicate evidence."
           },
           {
             "left": "pull_request:44003",
             "right": "pull_request:44004",
             "accept": false,
-            "reason": "Different codepaths (mamba vs codegen); only the refactor pattern overlaps."
+            "reason": "Different model internals and output shapes; not the same fix."
           },
           {
             "left": "pull_request:43998",
             "right": "pull_request:44004",
             "accept": false,
-            "reason": "timm_backbone and codegen are unrelated; not the same bug/change."
+            "reason": "Backbone output tracing vs CodeGen cache/output tracing are different changes."
+          },
+          {
+            "left": "pull_request:44001",
+            "right": "pull_request:44004",
+            "accept": false,
+            "reason": "UnivNet and CodeGen are unrelated model code paths."
+          },
+          {
+            "left": "pull_request:44002",
+            "right": "pull_request:44004",
+            "accept": false,
+            "reason": "UperNet segmentation and CodeGen generation logic are not the same bug."
+          },
+          {
+            "left": "pull_request:44000",
+            "right": "pull_request:44004",
+            "accept": false,
+            "reason": "Different models and output contracts; only broad refactor similarity."
           },
           {
             "left": "pull_request:44025",
             "right": "pull_request:44026",
             "accept": false,
-            "reason": "depth_anything/prompt_depth_anything and vision_encoder_decoder are different models; reject."
+            "reason": "Depth Anything vs Vision Encoder-Decoder are separate model changes."
           },
           {
             "left": "pull_request:44025",
             "right": "pull_request:44027",
             "accept": false,
-            "reason": "Different model families; shared helper import is not enough to duplicate them."
+            "reason": "Different encoder-decoder stacks; not one concrete fix."
           },
           {
             "left": "pull_request:44025",
             "right": "pull_request:44028",
             "accept": false,
-            "reason": "depth_anything vs superpoint are unrelated concrete codepaths."
+            "reason": "Depth estimation vs keypoint model; unrelated code paths."
           },
           {
             "left": "pull_request:44025",
             "right": "pull_request:44029",
             "accept": false,
-            "reason": "Different model families; same output-tracing refactor theme only."
+            "reason": "Different model families; same tracing refactor pattern only."
           },
           {
             "left": "pull_request:44025",
             "right": "pull_request:44030",
             "accept": false,
-            "reason": "depth_anything and dpr are different implementations; not a duplicate PR."
+            "reason": "Depth Anything and DPR have distinct implementations and behavior."
           },
           {
             "left": "pull_request:44026",
-            "right": "pull_request:44027",
+            "right": "pull_request:44029",
             "accept": false,
-            "reason": "vision_encoder_decoder and speech_encoder_decoder are related wrapper types, but the actual codepaths and change contexts differ; too broad to treat as one duplicate."
+            "reason": "Vision encoder-decoder and RWKV are unrelated model code paths."
           },
           {
             "left": "pull_request:44026",
-            "right": "pull_request:44028",
+            "right": "pull_request:44027",
             "accept": false,
-            "reason": "Different model families and files; reject."
+            "reason": "Different encoder-decoder models; too broad to merge as duplicates."
           },
           {
             "left": "pull_request:44026",
-            "right": "pull_request:44029",
+            "right": "pull_request:44028",
             "accept": false,
-            "reason": "Different models (vision_encoder_decoder vs rwkv); not the same change."
+            "reason": "Distinct model architectures and outputs; not duplicate work."
           },
           {
             "left": "pull_request:44026",
             "right": "pull_request:44030",
             "accept": false,
-            "reason": "Different codepaths and model architectures; only superficial similarity."
+            "reason": "Different model classes and code paths; only the refactor theme overlaps."
           },
           {
             "left": "pull_request:44027",
             "right": "pull_request:44028",
             "accept": false,
-            "reason": "speech_encoder_decoder and superpoint are unrelated model implementations."
+            "reason": "Speech encoder-decoder vs SuperPoint are not the same underlying change."
           },
           {
             "left": "pull_request:44027",
             "right": "pull_request:44029",
             "accept": false,
-            "reason": "Different model families; same tracing refactor pattern only."
+            "reason": "Unrelated model internals; shared tracing refactor is insufficient."
           },
           {
             "left": "pull_request:44027",
             "right": "pull_request:44030",
             "accept": false,
-            "reason": "speech_encoder_decoder vs dpr are different codepaths; not a duplicate."
+            "reason": "Different model families and behavior; not a duplicate."
           },
           {
             "left": "pull_request:44028",
             "right": "pull_request:44029",
             "accept": false,
-            "reason": "Different model families (superpoint vs rwkv); not the same concrete bug/change."
+            "reason": "SuperPoint and RWKV are separate subsystems with no shared concrete fix."
           },
           {
             "left": "pull_request:44028",
             "right": "pull_request:44030",
             "accept": false,
-            "reason": "Different implementations and change contexts; reject."
+            "reason": "Different model implementations; no same code-path issue."
           },
           {
             "left": "pull_request:44029",
             "right": "pull_request:44030",
             "accept": false,
-            "reason": "rwkv and dpr are different model families; shared refactor motif is insufficient for duplicate triage."
+            "reason": "RWKV and DPR touch different architectures and output handling."
           }
         ]
       },
@@ -18281,8 +18311,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "cache",
-      "cache_hit": true,
+      "source": "llm",
+      "cache_hit": false,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-44662-3",
       "nodes": [
@@ -18295,52 +18325,52 @@
         "pull_request:44662|pull_request:45075",
         "pull_request:45075|pull_request:45401"
       ],
-      "prepared_review_unit_hash": "d87fb88bf306802163835b7f004498cd12f41f72dba25bb378663745e43b96aa",
+      "prepared_review_unit_hash": "1582ed2fc980c4e74af4b0db6df84c61d389c68623c8905a582892cfb7c903d4",
       "component_budget": {
         "node_count": 3,
         "item_count": 3,
         "soft_pair_count": 3,
-        "serialized_chars": 10913,
-        "estimated_input_tokens": 2729,
-        "estimated_eval_tokens": 5714
+        "serialized_chars": 10912,
+        "estimated_input_tokens": 2728,
+        "estimated_eval_tokens": 5712
       },
       "budget": {
         "node_count": 3,
         "item_count": 3,
         "soft_pair_count": 3,
-        "serialized_chars": 10913,
-        "estimated_input_tokens": 2729,
-        "estimated_eval_tokens": 5714
+        "serialized_chars": 10912,
+        "estimated_input_tokens": 2728,
+        "estimated_eval_tokens": 5712
       },
       "overflow_policy": "truncate_then_skip",
       "trimmed": false,
       "aggressively_trimmed": false,
       "split": false,
       "analyst_result": {
-        "summary": "Three open PRs for adding different models (PenguinVL, DeepSeek-OCR-2, Voxtral-TTS). They share transformer scaffolding and docs/auto-registration changes, but they are distinct model implementations, not the same underlying change.",
-        "confidence": 0.95,
+        "summary": "These are three separate model-addition PRs (PenguinVL, DeepSeek-OCR-2, and Voxtral TTS) that share the same Transformers boilerplate but do not fix or implement the same underlying change.",
+        "confidence": 0.96,
         "canonical_issue_reason": null,
-        "canonical_pr_reason": "PR 45075 is the best representative if one is needed: it is a full model add with an explicit issue target, substantial implementation/test coverage, and broader repository integration. It is still not a duplicate of the other PRs.",
+        "canonical_pr_reason": "No canonical PR: each PR introduces a different model integration with distinct configs, processors, model code, and tests.",
         "best_issue_reason": null,
-        "best_pr_reason": "PR 45075 is the strongest standalone PR in the set because it is the most anchored and complete, but the cluster does not contain duplicate PRs.",
+        "best_pr_reason": "None stands out as a duplicate representative; the closest match is not a real duplicate because the PRs target different models and code paths.",
         "soft_edge_verdicts": [
           {
             "left": "pull_request:44662",
             "right": "pull_request:45401",
             "accept": false,
-            "reason": "Different model additions: PenguinVL vs Voxtral TTS. Overlap is only in shared registration/docs scaffolding, not the same code-path bug or change."
+            "reason": "Different models and different implementations (PenguinVL vs Voxtral TTS); shared auto/docs files are just standard model-registration boilerplate, not the same fix."
           },
           {
             "left": "pull_request:44662",
             "right": "pull_request:45075",
             "accept": false,
-            "reason": "Different models and different implementation goals. Shared auto-mapping and docs files are broad integration points, not evidence of the same underlying change."
+            "reason": "Both add new model support, but for different models (PenguinVL vs DeepSeek-OCR-2) with separate architecture, conversion, and test code."
           },
           {
             "left": "pull_request:45075",
             "right": "pull_request:45401",
             "accept": false,
-            "reason": "DeepSeek-OCR-2 and Voxtral-TTS are separate model implementations with different files and behaviors; they only coincide on generic transformers boilerplate."
+            "reason": "Distinct model additions with different target problems and unique source trees; not mergeable as one PR for the same underlying change."
           }
         ]
       },
@@ -18359,8 +18389,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "cache",
-      "cache_hit": true,
+      "source": "llm",
+      "cache_hit": false,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-44979-2",
       "nodes": [
@@ -18370,12 +18400,12 @@
       "soft_pairs": [
         "pull_request:44979|pull_request:45363"
       ],
-      "prepared_review_unit_hash": "e6b38a2f06c83d53669bfbae77b45122e42f98f00100f0b8a0c6e60198bc070f",
+      "prepared_review_unit_hash": "db0324771111ae121e3998f1e5cbb1f1eac6293592181e1ad89e5f4517728939",
       "component_budget": {
         "node_count": 2,
         "item_count": 2,
         "soft_pair_count": 1,
-        "serialized_chars": 4407,
+        "serialized_chars": 4406,
         "estimated_input_tokens": 1102,
         "estimated_eval_tokens": 2460
       },
@@ -18383,7 +18413,7 @@
         "node_count": 2,
         "item_count": 2,
         "soft_pair_count": 1,
-        "serialized_chars": 4407,
+        "serialized_chars": 4406,
         "estimated_input_tokens": 1102,
         "estimated_eval_tokens": 2460
       },
@@ -18392,18 +18422,18 @@
       "aggressively_trimmed": false,
       "split": false,
       "analyst_result": {
-        "summary": "The two PRs are related to fusion infrastructure, but they are not the same change: #44979 introduces a new module-fusion API, while #45363 adds n-to-1 kernel fusion wiring via KernelConfig and integration changes. They overlap conceptually, but not as duplicate fixes or mergeable one-PR equivalents.",
+        "summary": "Both PRs are about fusion-related APIs in Transformers, but they target different layers: #44979 adds a generic module-fusion API, while #45363 implements n-to-1 kernel fusion via `KernelConfig` and related integration changes. They are related, but not the same underlying change.",
         "confidence": 0.89,
         "canonical_issue_reason": null,
-        "canonical_pr_reason": "#44979 is the cleaner canonical representative of the cluster because it defines the core Module Fusion API itself, whereas #45363 is a more specific KernelConfig-based integration on top of that idea.",
+        "canonical_pr_reason": "#45363 is the more complete, user-facing fusion feature: it wires `KernelConfig` through model loading/integrations and includes an end-to-end example. #44979 is a separate foundational module-fusion API addition, not the same concrete change.",
         "best_issue_reason": null,
-        "best_pr_reason": "#44979 is the best single PR to represent the cluster theme: it is self-contained, foundational, and directly about the fusion abstraction rather than an example or integration layer.",
+        "best_pr_reason": "#45363 is the better representative of the cluster because it contains the broader concrete implementation path for kernel fusion, touching core loading/integration code rather than only introducing a standalone API surface.",
         "soft_edge_verdicts": [
           {
             "left": "pull_request:44979",
             "right": "pull_request:45363",
             "accept": false,
-            "reason": "Related fusion work, but different underlying changes: one adds a module-fusion API, the other implements kernel-config-driven n-to-1 fusion integration. They do not look like the same bug or change."
+            "reason": "Related fusion work, but not the same bug or change. #44979 adds a generic module-fusion API in a new file, while #45363 implements kernel fusion through `KernelConfig` and integration changes. They would not plausibly be merged as one PR without collapsing distinct features."
           }
         ]
       },
@@ -18422,8 +18452,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "cache",
-      "cache_hit": true,
+      "source": "llm",
+      "cache_hit": false,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-45086-2",
       "nodes": [
@@ -18433,7 +18463,7 @@
       "soft_pairs": [
         "pull_request:45086|pull_request:45317"
       ],
-      "prepared_review_unit_hash": "85a47b91c5a3a66b4d0ad0ad165e8446ca0233204f9c165fb613d69f611435f2",
+      "prepared_review_unit_hash": "f75659896d3f7df0091b894c3bd8554367c49f17aa6da9486b53e9524e366814",
       "component_budget": {
         "node_count": 2,
         "item_count": 2,
@@ -18455,18 +18485,18 @@
       "aggressively_trimmed": false,
       "split": false,
       "analyst_result": {
-        "summary": "Both PRs implement the same fix for `_patch_mistral_regex` in `tokenization_utils_tokenizers.py`; 45317 is the more complete version because it adds tests, while 45086 is essentially the earlier minimal patch.",
-        "confidence": 0.97,
+        "summary": "Both pull requests fix the same AttributeError in `_patch_mistral_regex` by switching from `tokenizer.backend_tokenizer.pre_tokenizer` to `tokenizer.pre_tokenizer` in `tokenization_utils_tokenizers.py`; 45317 is the more complete version because it also adds tests.",
+        "confidence": 0.99,
         "canonical_issue_reason": null,
-        "canonical_pr_reason": "PR 45317 is the best canonical PR: it fixes the same AttributeError in the same code path as 45086 and also includes tests, making it the more complete representative of the change.",
+        "canonical_pr_reason": "45317 is the better canonical PR because it implements the same bug fix as 45086 and includes accompanying test coverage, making it the more complete representative of the change.",
         "best_issue_reason": null,
-        "best_pr_reason": "PR 45317 is the strongest overall PR candidate because it contains the same fix plus test coverage, whereas 45086 is just the minimal implementation patch.",
+        "best_pr_reason": "45317 is the best PR candidate because it addresses the same concrete Mistral pre-tokenizer AttributeError and adds validation tests, so it is the strongest standalone fix in the cluster.",
         "soft_edge_verdicts": [
           {
             "left": "pull_request:45086",
             "right": "pull_request:45317",
             "accept": true,
-            "reason": "Same underlying bug and same file/change: both replace `backend_tokenizer.pre_tokenizer` with `tokenizer.pre_tokenizer` in `_patch_mistral_regex` to fix the AttributeError. 45317 is a superset with tests, so these are mergeable as one PR."
+            "reason": "Same underlying code-path bug in `_patch_mistral_regex`, same file, same explicit issue target, and the diffs overlap on the same fix; 45317 just adds tests."
           }
         ]
       },
@@ -18485,8 +18515,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "cache",
-      "cache_hit": true,
+      "source": "llm",
+      "cache_hit": false,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-45115-2",
       "nodes": [
@@ -18496,7 +18526,7 @@
       "soft_pairs": [
         "pull_request:45115|pull_request:45569"
       ],
-      "prepared_review_unit_hash": "6cf91c564ea3f1ceb40b647300b54192eb5b689adfb50bb9a487006f4198ae5c",
+      "prepared_review_unit_hash": "e6750721eb8783fdf126499ec017eeb37c80aed37bd362e5012beeec4196f714",
       "component_budget": {
         "node_count": 2,
         "item_count": 2,
@@ -18518,18 +18548,18 @@
       "aggressively_trimmed": false,
       "split": false,
       "analyst_result": {
-        "summary": "Two open NemotronH PRs touch overlapping model code, but they address different changes: one is a refactor/conversion-mapping inheritance update, the other is a broader split into NemotronH dense/sparse variants with docs and tests. They are related but not duplicates.",
-        "confidence": 0.88,
+        "summary": "The two PRs overlap on NemotronH codepaths, but they are not the same change: #45115 is a narrower inheritance/conversion-mapping refactor toward GraniteMoeHybrid compatibility, while #45569 is a broader model-family split introducing NemotronHDense/NemotronHSparse plus docs and auto-registration updates.",
+        "confidence": 0.92,
         "canonical_issue_reason": null,
-        "canonical_pr_reason": "PR #45569 is the broader, more user-facing NemotronH architecture change and includes the new dense/sparse model surfaces plus docs/tests, so it is the better representative of the cluster.",
+        "canonical_pr_reason": "#45569 is the more complete and user-facing PR: it defines the new NemotronH dense/sparse variants, updates docs, and wires the model into auto/config/model registration. It better represents the main architectural change in this cluster.",
         "best_issue_reason": null,
-        "best_pr_reason": "PR #45569 is the strongest standalone candidate because it appears to implement the main intended model restructuring rather than just the supporting conversion/refactor work.",
+        "best_pr_reason": "#45569 is the best PR because it covers the full NemotronH refactor/splitting effort rather than only a conversion-mapping/inheritance adjustment.",
         "soft_edge_verdicts": [
           {
             "left": "pull_request:45115",
             "right": "pull_request:45569",
             "accept": false,
-            "reason": "They share NemotronH-related files, but #45115 is mainly a conversion-mapping/refactor to GraniteMoeHybrid-compatible names, while #45569 is a broader model split and documentation/test update. They do not look like the same concrete fix and would not plausibly merge into one PR."
+            "reason": "They share some NemotronH files, but the underlying changes differ: #45115 focuses on checkpoint conversion naming and GraniteMoeHybrid-style inheritance, whereas #45569 introduces separate dense/sparse model variants and documentation/auto-map updates. They are related, not duplicates."
           }
         ]
       },
@@ -18548,8 +18578,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "cache",
-      "cache_hit": true,
+      "source": "llm",
+      "cache_hit": false,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-45332-2",
       "nodes": [
@@ -18559,7 +18589,7 @@
       "soft_pairs": [
         "pull_request:45332|pull_request:45333"
       ],
-      "prepared_review_unit_hash": "1e292ea90364caf2eb05d1180da055756c14994a55bb1fe1678a127e360388af",
+      "prepared_review_unit_hash": "c252bdbcbd96340711e2188c1c881d12f27b95eb3952546a91bf31e460b86297",
       "component_budget": {
         "node_count": 2,
         "item_count": 2,
@@ -18581,18 +18611,18 @@
       "aggressively_trimmed": false,
       "split": false,
       "analyst_result": {
-        "summary": "These are related feature PRs, but not duplicates: one adds heterogeneous per-layer config support, the other adds broader heterogeneous modeling/cache behavior built on top of that config layer.",
-        "confidence": 0.84,
+        "summary": "Two open PRs propose related heterogeneous configuration/modeling work, but they are not duplicates: one adds per-layer config support, the other adds broader heterogeneous model support that builds on it. The larger PR is the better cluster representative.",
+        "confidence": 0.89,
         "canonical_issue_reason": null,
-        "canonical_pr_reason": "PR #45332 is the broader umbrella change: it implements the actual heterogeneous model/runtime behavior, not just the config plumbing.",
+        "canonical_pr_reason": "PR #45332 is the broader, more central change: it covers per-layer config plus the modeling/cache/masking updates that use it. PR #45333 is narrower and largely foundational/config-only.",
         "best_issue_reason": null,
-        "best_pr_reason": "PR #45332 is the best representative for the cluster because it covers the end-to-end heterogeneous support path and subsumes the smaller config-only foundation.",
+        "best_pr_reason": "PR #45332 is the best representative of the cluster because it subsumes the shared heterogeneity groundwork and extends into the main runtime/modeling path, making it the more complete canonical PR.",
         "soft_edge_verdicts": [
           {
             "left": "pull_request:45332",
             "right": "pull_request:45333",
             "accept": false,
-            "reason": "They are adjacent pieces of the same feature area, but not the same concrete change: #45333 adds per-layer config plumbing, while #45332 adds modeling/cache behavior using that plumbing. They would not plausibly be merged as one duplicate PR."
+            "reason": "Related feature work, but not the same concrete change or bug fix. #45333 focuses on per-layer configuration support, while #45332 adds broader heterogeneous model/runtime support that consumes that config. They are complementary rather than duplicate PRs."
           }
         ]
       },
@@ -18611,8 +18641,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "cache",
-      "cache_hit": true,
+      "source": "llm",
+      "cache_hit": false,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-45415-2",
       "nodes": [
@@ -18622,7 +18652,7 @@
       "soft_pairs": [
         "pull_request:45415|pull_request:45425"
       ],
-      "prepared_review_unit_hash": "3b99fdb7a22c2a9ae34be195a477e6fbaeeb2f8412d798db27d6a28546635d16",
+      "prepared_review_unit_hash": "16137db7fa18870577d38ab41de5da181f80b0decfeaa00210acd97424fae61b",
       "component_budget": {
         "node_count": 2,
         "item_count": 2,
@@ -18644,18 +18674,18 @@
       "aggressively_trimmed": false,
       "split": false,
       "analyst_result": {
-        "summary": "These are both typing-related PRs, but they address different changes: #45415 is a broad repo-wide type-checking cleanup, while #45425 adds specific typing protocols/aliases and related fixes. They are similar in theme, not duplicates of the same underlying change.",
-        "confidence": 0.88,
+        "summary": "Two typing-focused PRs overlap in a few files, but they address different scopes: one is a broad type-checking sweep across many transformers modules, and the other adds specific typing protocols/helpers for modeling and PEFT-related code paths.",
+        "confidence": 0.82,
         "canonical_issue_reason": null,
-        "canonical_pr_reason": "PR #45425 is the more focused, concrete typing-infra change (new aliases/protocols and targeted fixes), whereas #45415 is a broad type-checking sweep across many files.",
+        "canonical_pr_reason": "PR #45425 is the more focused and concrete change: it adds shared typing aliases/protocols and updates a small set of code paths to use them. PR #45415 is a broader repository-wide typing cleanup/configuration pass, so it is less representative as the cluster anchor.",
         "best_issue_reason": null,
-        "best_pr_reason": "PR #45425 is the better representative of the cluster because its scope is narrower and more self-contained, making it the clearest single change to track.",
+        "best_pr_reason": "PR #45425 is the best representative for this cluster because it centers on a specific typing abstraction change rather than a wide, multi-file type-checking sweep. PR #45415 looks like a larger housekeeping effort with many unrelated edits, making it a weaker canonical representative.",
         "soft_edge_verdicts": [
           {
             "left": "pull_request:45415",
             "right": "pull_request:45425",
             "accept": false,
-            "reason": "Both concern typing, but #45415 is a large general type-checking pass and #45425 is a specific set of typing protocol/alias additions; they do not look like the same concrete fix and would not plausibly be merged as one PR."
+            "reason": "Do not merge: the overlap is only superficial typing-related cleanup. #45415 is a broad type-checking pass across many files and pyproject settings, while #45425 introduces specific typing protocols and helpers for modeling/PEFT code paths. They do not appear to fix the same concrete bug or change."
           }
         ]
       },
@@ -18690,7 +18720,7 @@
         "pull_request:45817|pull_request:45843",
         "pull_request:45836|pull_request:45843"
       ],
-      "prepared_review_unit_hash": "9b112161716be4895913ffb98ae3da14ed8dbadc7448fbe67a752e9cbc4f0add",
+      "prepared_review_unit_hash": "6a96123a7dd219e95f907ba16cbb534a26673e315d9e8a31962abce02616020b",
       "component_budget": {
         "node_count": 4,
         "item_count": 4,
@@ -18712,36 +18742,36 @@
       "aggressively_trimmed": false,
       "split": false,
       "analyst_result": {
-        "summary": "These pull requests are thematically similar around model-parallel or test-stability work, but they target different models and different concrete failures, so they should not be deduplicated.",
-        "confidence": 0.93,
+        "summary": "This cluster is mostly a set of unrelated model-parallel/device-placement fixes across different model families, plus one unrelated test-skip PR. They are thematically similar but not the same underlying change.",
+        "confidence": 0.91,
         "canonical_issue_reason": null,
-        "canonical_pr_reason": "PR 45817 is the clearest representative code fix: it addresses an actual model-parallel bug in Gemma4 with focused implementation changes and tests, rather than being mostly test skips or a broader multi-model sweep.",
+        "canonical_pr_reason": "pull_request:45487 is the broadest and clearest representative of the shared model-parallel/device-placement theme, with concrete code changes rather than only test adjustments.",
         "best_issue_reason": null,
-        "best_pr_reason": "PR 45817 is the best standalone PR in this cluster because it fixes a specific code-path problem in the model itself and is well-scoped enough to stand as a representative bug fix.",
+        "best_pr_reason": "pull_request:45487 best captures the cluster\u2019s main pattern of fixing model-parallel placement issues in model code; the others are either narrower, model-specific variants or unrelated test skips.",
         "soft_edge_verdicts": [
           {
             "left": "pull_request:45487",
             "right": "pull_request:45843",
             "accept": false,
-            "reason": "Both are model-parallel themed, but 45487 fixes token_type_ids device handling in AltCLIP/ChineseCLIP while 45843 propagates _no_split_modules in Deimv2. Different models and different failure modes."
+            "reason": "Both address model-parallel placement, but they fix different model families and different code paths; not the same concrete bug or mergeable as one PR."
           },
           {
             "left": "pull_request:45487",
             "right": "pull_request:45836",
             "accept": false,
-            "reason": "45487 is a real code fix; 45836 only skips failing tests for MiniCPM-V 4.6/Qwen3.5 cache compatibility. Not the same underlying bug or change."
+            "reason": "45487 fixes runtime device-placement/model-parallel behavior, while 45836 only skips failing tests for a cache incompatibility; unrelated underlying change."
           },
           {
             "left": "pull_request:45817",
             "right": "pull_request:45843",
             "accept": false,
-            "reason": "Gemma4 model-parallel fixes and Deimv2 model-parallel fixes are different concrete bugs in different code paths, so they are not duplicates."
+            "reason": "Similar high-level theme, but Gemma4 and Deimv2 have distinct implementations and fixes; not the same underlying bug."
           },
           {
             "left": "pull_request:45836",
             "right": "pull_request:45843",
             "accept": false,
-            "reason": "One PR skips incompatible tests, the other changes Deimv2 model-parallel behavior in production code. These are not the same underlying issue."
+            "reason": "Test-only skip for MiniCPM/Qwen cache incompatibility versus a Deimv2 model-parallel code fix; clearly different issues."
           }
         ]
       },
@@ -18760,8 +18790,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "cache",
-      "cache_hit": true,
+      "source": "llm",
+      "cache_hit": false,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-45548-2",
       "nodes": [
@@ -18771,7 +18801,7 @@
       "soft_pairs": [
         "pull_request:45548|pull_request:45662"
       ],
-      "prepared_review_unit_hash": "7b7d08b43c9d5c8b4ea8de299d8d2982907f722c2ecfcc46a1a6f6763c0d87de",
+      "prepared_review_unit_hash": "70c5056f3f907c8ae594b89749e73943c9328affc7a2cee66a9e02228aa970c6",
       "component_budget": {
         "node_count": 2,
         "item_count": 2,
@@ -18793,18 +18823,18 @@
       "aggressively_trimmed": false,
       "split": false,
       "analyst_result": {
-        "summary": "Both PRs are about EP loading, but they address different concrete failures: PR #45548 fixes config/accelerate + DeepSpeed ZeRO-3 loading, while PR #45662 fixes expert tensors being overwritten during FSDP2/rank-0 broadcast. Shared subsystem is not enough to treat them as duplicates.",
-        "confidence": 0.97,
+        "summary": "These two PRs both touch expert-parallel loading, but they address different bugs and code paths: one fixes EP + ZeRO-3/accelerate launch config handling, the other fixes EP + FSDP2 expert tensors being overwritten by rank-0 broadcast.",
+        "confidence": 0.94,
         "canonical_issue_reason": null,
-        "canonical_pr_reason": "No canonical PR: the two patches fix different code-path bugs in EP loading and should not be merged as one change.",
+        "canonical_pr_reason": "No clear canonical duplicate here; PR 45662 is the more self-contained fix for a specific overwrite bug in distributed loading, while 45548 targets a separate ZeRO-3/config serialization issue.",
         "best_issue_reason": null,
-        "best_pr_reason": "If forced to pick a representative, PR #45662 is the broader EP-loading fix, but it is still not a duplicate of PR #45548.",
+        "best_pr_reason": "PR 45662 is the stronger representative of a concrete, narrowly scoped bug fix with multiple code-path changes, but it should not be treated as a duplicate of 45548.",
         "soft_edge_verdicts": [
           {
             "left": "pull_request:45548",
             "right": "pull_request:45662",
             "accept": false,
-            "reason": "Different underlying bugs and fix paths: #45548 preserves distributed_config for accelerate/ZeRO-3 loading; #45662 prevents expert parameters from being overwritten via post-shard wrapping under FSDP2/broadcast."
+            "reason": "Shared EP/distributed-loading theme is too broad; 45548 fixes accelerate/ZeRO-3 config handling and `distributed_config` serialization, while 45662 fixes FSDP2 rank-0 broadcast overwriting experts via `post_shard_wrap` and related loading paths."
           }
         ]
       },
@@ -18823,8 +18853,8 @@
       "review_unit_count": 1,
       "status": "reviewed",
       "reason": null,
-      "source": "cache",
-      "cache_hit": true,
+      "source": "llm",
+      "cache_hit": false,
       "model": "gpt-5.4-mini?service_tier=flex",
       "cluster_id": "cluster-45639-2",
       "nodes": [
@@ -18834,12 +18864,12 @@
       "soft_pairs": [
         "pull_request:45639|pull_request:45645"
       ],
-      "prepared_review_unit_hash": "0f219a20c7e41a3eef2bde62c50aef1f52842d773c7fc5415276397f7f013561",
+      "prepared_review_unit_hash": "7bdffe171c433d63389d9812273698bf553b4f96b3aa95218a39f09d39882fc2",
       "component_budget": {
         "node_count": 2,
         "item_count": 2,
         "soft_pair_count": 1,
-        "serialized_chars": 4590,
+        "serialized_chars": 4589,
         "estimated_input_tokens": 1148,
         "estimated_eval_tokens": 2552
       },
@@ -18847,7 +18877,7 @@
         "node_count": 2,
         "item_count": 2,
         "soft_pair_count": 1,
-        "serialized_chars": 4590,
+        "serialized_chars": 4589,
         "estimated_input_tokens": 1148,
         "estimated_eval_tokens": 2552
       },
@@ -18856,18 +18886,18 @@
       "aggressively_trimmed": false,
       "split": false,
       "analyst_result": {
-        "summary": "Two overlapping PRs address the same xdist-safe captured_info logging bug; one is a narrower code-only fix, the other extends it to CI log collection and is the more complete patch.",
-        "confidence": 0.94,
+        "summary": "Both PRs target the same xdist-related captured_info logging problem; PR 45645 appears to subsume PR 45639 with the same testing_utils fix plus the CI artifact handling changes.",
+        "confidence": 0.89,
         "canonical_issue_reason": null,
-        "canonical_pr_reason": "PR #45645 is the broader, more complete implementation of the captured_info/xdist collision fix and includes the CI-side handling needed to preserve logs.",
+        "canonical_pr_reason": "PR 45645 is the broader, more complete fix for the captured_info xdist collision issue: it includes the testing_utils change plus workflow/notification updates to handle multiple captured_info files in CI.",
         "best_issue_reason": null,
-        "best_pr_reason": "PR #45645 best represents the cluster because it fixes the core bug and also updates the workflow that consumes the generated artifacts.",
+        "best_pr_reason": "PR 45645 is the best representative because it addresses the underlying bug end-to-end, while 45639 is a narrower subset limited to making the debug log write path xdist-safe.",
         "soft_edge_verdicts": [
           {
             "left": "pull_request:45639",
             "right": "pull_request:45645",
             "accept": true,
-            "reason": "Both PRs target the same underlying captured_info xdist collision problem in testing_utils; #45645 is a superset of the same fix and could plausibly subsume #45639 into one PR."
+            "reason": "Same underlying xdist/captured_info collision problem, same issue target, and 45645 appears to extend rather than diverge from 45639; these could plausibly be merged into one PR."
           }
         ]
       },
diff --git a/analysis/current/manifest.json b/analysis/current/manifest.json
index 95796a666fe7d739c40a7bd540801811610f42b1..40cbeec7407eac6061ca3583f7fe177051867ddc 100644
--- a/analysis/current/manifest.json
+++ b/analysis/current/manifest.json
@@ -1,8 +1,8 @@
 {
-  "analysis_id": "hybrid-model-20260510t000023z",
+  "analysis_id": "hybrid-model-20260510t060019z",
   "archived_artifacts": {
-    "hybrid": "snapshots/20260510T000023Z/analysis-runs/hybrid-model-20260510t000023z/analysis-report-hybrid.json",
-    "hybrid_reviews": "snapshots/20260510T000023Z/analysis-runs/hybrid-model-20260510t000023z/analysis-report-hybrid.llm-reviews.json"
+    "hybrid": "snapshots/20260510T060019Z/analysis-runs/hybrid-model-20260510t060019z/analysis-report-hybrid.json",
+    "hybrid_reviews": "snapshots/20260510T060019Z/analysis-runs/hybrid-model-20260510t060019z/analysis-report-hybrid.llm-reviews.json"
   },
   "artifacts": {
     "hybrid": "analysis/current/analysis-report-hybrid.json",
@@ -10,9 +10,9 @@
   },
   "channel": "canonical",
   "model": null,
-  "published_at": "2026-05-10T00:05:57Z",
+  "published_at": "2026-05-10T06:03:19Z",
   "repo": "huggingface/transformers",
   "schema_version": 1,
-  "snapshot_id": "20260510T000023Z",
+  "snapshot_id": "20260510T060019Z",
   "variant": "hybrid"
 }
diff --git a/snapshots/20260510T060019Z/analysis-runs/hybrid-model-20260510t060019z/analysis-report-hybrid.json b/snapshots/20260510T060019Z/analysis-runs/hybrid-model-20260510t060019z/analysis-report-hybrid.json
new file mode 100644
index 0000000000000000000000000000000000000000..886eae36670f9bc7a423b6b301ff1618a2deb775
--- /dev/null
+++ b/snapshots/20260510T060019Z/analysis-runs/hybrid-model-20260510t060019z/analysis-report-hybrid.json
@@ -0,0 +1,2076 @@
+{
+  "schema_version": "1.0",
+  "repo": "huggingface/transformers",
+  "snapshot_id": "20260510T060019Z",
+  "generated_at": "2026-05-10T06:03:17Z",
+  "evidence_quality": "full",
+  "llm_enrichment": true,
+  "meta_bugs": [
+    {
+      "cluster_id": "cluster-43979-11",
+      "summary": "Cluster of 1 issues and 10 PRs centered on issue #43979.",
+      "status": "open",
+      "confidence": 0.8,
+      "canonical_issue_number": 43979,
+      "canonical_pr_number": 44007,
+      "issue_numbers": [
+        43979
+      ],
+      "pr_numbers": [
+        43996,
+        44007,
+        44013,
+        44044,
+        44066,
+        44072,
+        44085,
+        44129,
+        44154,
+        44722
+      ],
+      "evidence_types": [
+        "closing_reference",
+        "shared_issue_target",
+        "soft_similarity"
+      ],
+      "pr_comparisons": [
+        {
+          "left_pr_number": 43996,
+          "right_pr_number": 44007,
+          "code_similarity": 0.179,
+          "size_similarity": 0.576,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.429,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 43996,
+          "right_pr_number": 44013,
+          "code_similarity": 0.122,
+          "size_similarity": 0.318,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.392,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 43996,
+          "right_pr_number": 44044,
+          "code_similarity": 0.245,
+          "size_similarity": 0.864,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.479,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 43996,
+          "right_pr_number": 44066,
+          "code_similarity": 0.225,
+          "size_similarity": 0.818,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.408,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 43996,
+          "right_pr_number": 44072,
+          "code_similarity": 0.14,
+          "size_similarity": 0.303,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.528,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 43996,
+          "right_pr_number": 44085,
+          "code_similarity": 0.216,
+          "size_similarity": 0.783,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.398,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 43996,
+          "right_pr_number": 44129,
+          "code_similarity": 0.163,
+          "size_similarity": 0.643,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.229,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 43996,
+          "right_pr_number": 44154,
+          "code_similarity": 0.153,
+          "size_similarity": 0.535,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.31,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 43996,
+          "right_pr_number": 44722,
+          "code_similarity": 0.225,
+          "size_similarity": 0.848,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.368,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44007,
+          "right_pr_number": 44013,
+          "code_similarity": 0.19,
+          "size_similarity": 0.553,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.531,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44007,
+          "right_pr_number": 44044,
+          "code_similarity": 0.186,
+          "size_similarity": 0.667,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.354,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44007,
+          "right_pr_number": 44066,
+          "code_similarity": 0.188,
+          "size_similarity": 0.704,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.315,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44007,
+          "right_pr_number": 44072,
+          "code_similarity": 0.212,
+          "size_similarity": 0.526,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.708,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44007,
+          "right_pr_number": 44085,
+          "code_similarity": 0.195,
+          "size_similarity": 0.735,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.318,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44007,
+          "right_pr_number": 44129,
+          "code_similarity": 0.103,
+          "size_similarity": 0.37,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.191,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44007,
+          "right_pr_number": 44154,
+          "code_similarity": 0.238,
+          "size_similarity": 0.93,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.344,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44007,
+          "right_pr_number": 44722,
+          "code_similarity": 0.178,
+          "size_similarity": 0.679,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.28,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44013,
+          "right_pr_number": 44044,
+          "code_similarity": 0.126,
+          "size_similarity": 0.368,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.351,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44013,
+          "right_pr_number": 44066,
+          "code_similarity": 0.127,
+          "size_similarity": 0.389,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.325,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44013,
+          "right_pr_number": 44072,
+          "code_similarity": 0.29,
+          "size_similarity": 0.952,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.667,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44013,
+          "right_pr_number": 44085,
+          "code_similarity": 0.131,
+          "size_similarity": 0.406,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.329,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44013,
+          "right_pr_number": 44129,
+          "code_similarity": 0.07,
+          "size_similarity": 0.205,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.192,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44013,
+          "right_pr_number": 44154,
+          "code_similarity": 0.177,
+          "size_similarity": 0.594,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.389,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44013,
+          "right_pr_number": 44722,
+          "code_similarity": 0.118,
+          "size_similarity": 0.375,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.287,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44044,
+          "right_pr_number": 44066,
+          "code_similarity": 0.25,
+          "size_similarity": 0.947,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.404,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44044,
+          "right_pr_number": 44072,
+          "code_similarity": 0.136,
+          "size_similarity": 0.351,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.442,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44044,
+          "right_pr_number": 44085,
+          "code_similarity": 0.24,
+          "size_similarity": 0.906,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.394,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44044,
+          "right_pr_number": 44129,
+          "code_similarity": 0.147,
+          "size_similarity": 0.555,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.243,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44044,
+          "right_pr_number": 44154,
+          "code_similarity": 0.17,
+          "size_similarity": 0.62,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.306,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44044,
+          "right_pr_number": 44722,
+          "code_similarity": 0.257,
+          "size_similarity": 0.982,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.402,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44066,
+          "right_pr_number": 44072,
+          "code_similarity": 0.133,
+          "size_similarity": 0.37,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.393,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44066,
+          "right_pr_number": 44085,
+          "code_similarity": 0.763,
+          "size_similarity": 0.957,
+          "file_overlap": 0.5,
+          "area_overlap": 0.825,
+          "patch_similarity": 0.887,
+          "shared_filenames": [
+            "src/transformers/models/gptj/modeling_gptj.py"
+          ],
+          "shared_file_areas": [
+            {
+              "filename": "src/transformers/models/gptj/modeling_gptj.py",
+              "left_ranges": [
+                [
+                  33,
+                  41
+                ],
+                [
+                  174,
+                  181
+                ],
+                [
+                  250,
+                  257
+                ],
+                [
+                  398,
+                  419
+                ],
+                [
+                  425,
+                  434
+                ],
+                [
+                  459,
+                  466
+                ],
+                [
+                  471,
+                  488
+                ],
+                [
+                  518,
+                  540
+                ],
+                [
+                  554,
+                  560
+                ],
+                [
+                  566,
+                  575
+                ],
+                [
+                  580,
+                  598
+                ],
+                [
+                  601,
+                  606
+                ],
+                [
+                  634,
+                  640
+                ],
+                [
+                  646,
+                  653
+                ],
+                [
+                  658,
+                  674
+                ],
+                [
+                  717,
+                  722
+                ],
+                [
+                  738,
+                  744
+                ],
+                [
+                  749,
+                  772
+                ],
+                [
+                  790,
+                  795
+                ]
+              ],
+              "right_ranges": [
+                [
+                  33,
+                  40
+                ],
+                [
+                  173,
+                  180
+                ],
+                [
+                  249,
+                  256
+                ],
+                [
+                  397,
+                  405
+                ],
+                [
+                  408,
+                  419
+                ],
+                [
+                  425,
+                  434
+                ],
+                [
+                  459,
+                  465
+                ],
+                [
+                  470,
+                  489
+                ],
+                [
+                  517,
+                  539
+                ],
+                [
+                  553,
+                  559
+                ],
+                [
+                  565,
+                  574
+                ],
+                [
+                  579,
+                  597
+                ],
+                [
+                  600,
+                  611
+                ],
+                [
+                  633,
+                  639
+                ],
+                [
+                  645,
+                  652
+                ],
+                [
+                  657,
+                  673
+                ],
+                [
+                  716,
+                  728
+                ],
+                [
+                  737,
+                  743
+                ],
+                [
+                  748,
+                  771
+                ],
+                [
+                  789,
+                  794
+                ]
+              ]
+            }
+          ]
+        },
+        {
+          "left_pr_number": 44066,
+          "right_pr_number": 44129,
+          "code_similarity": 0.145,
+          "size_similarity": 0.526,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.263,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44066,
+          "right_pr_number": 44154,
+          "code_similarity": 0.174,
+          "size_similarity": 0.654,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.286,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44066,
+          "right_pr_number": 44722,
+          "code_similarity": 0.907,
+          "size_similarity": 0.964,
+          "file_overlap": 1.0,
+          "area_overlap": 0.808,
+          "patch_similarity": 0.874,
+          "shared_filenames": [
+            "src/transformers/models/codegen/modeling_codegen.py",
+            "src/transformers/models/gptj/modeling_gptj.py"
+          ],
+          "shared_file_areas": [
+            {
+              "filename": "src/transformers/models/codegen/modeling_codegen.py",
+              "left_ranges": [
+                [
+                  245,
+                  266
+                ]
+              ],
+              "right_ranges": [
+                [
+                  228,
+                  234
+                ]
+              ]
+            },
+            {
+              "filename": "src/transformers/models/gptj/modeling_gptj.py",
+              "left_ranges": [
+                [
+                  33,
+                  41
+                ],
+                [
+                  174,
+                  181
+                ],
+                [
+                  250,
+                  257
+                ],
+                [
+                  398,
+                  419
+                ],
+                [
+                  425,
+                  434
+                ],
+                [
+                  459,
+                  466
+                ],
+                [
+                  471,
+                  488
+                ],
+                [
+                  518,
+                  540
+                ],
+                [
+                  554,
+                  560
+                ],
+                [
+                  566,
+                  575
+                ],
+                [
+                  580,
+                  598
+                ],
+                [
+                  601,
+                  606
+                ],
+                [
+                  634,
+                  640
+                ],
+                [
+                  646,
+                  653
+                ],
+                [
+                  658,
+                  674
+                ],
+                [
+                  717,
+                  722
+                ],
+                [
+                  738,
+                  744
+                ],
+                [
+                  749,
+                  772
+                ],
+                [
+                  790,
+                  795
+                ]
+              ],
+              "right_ranges": [
+                [
+                  33,
+                  42
+                ],
+                [
+                  175,
+                  182
+                ],
+                [
+                  251,
+                  258
+                ],
+                [
+                  399,
+                  420
+                ],
+                [
+                  426,
+                  435
+                ],
+                [
+                  460,
+                  467
+                ],
+                [
+                  472,
+                  489
+                ],
+                [
+                  519,
+                  540
+                ],
+                [
+                  554,
+                  560
+                ],
+                [
+                  566,
+                  575
+                ],
+                [
+                  580,
+                  598
+                ],
+                [
+                  601,
+                  612
+                ],
+                [
+                  634,
+                  640
+                ],
+                [
+                  646,
+                  653
+                ],
+                [
+                  658,
+                  663
+                ],
+                [
+                  666,
+                  674
+                ],
+                [
+                  717,
+                  722
+                ],
+                [
+                  738,
+                  744
+                ],
+                [
+                  749,
+                  772
+                ],
+                [
+                  790,
+                  795
+                ]
+              ]
+            }
+          ]
+        },
+        {
+          "left_pr_number": 44072,
+          "right_pr_number": 44085,
+          "code_similarity": 0.137,
+          "size_similarity": 0.387,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.398,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44072,
+          "right_pr_number": 44129,
+          "code_similarity": 0.074,
+          "size_similarity": 0.195,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.231,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44072,
+          "right_pr_number": 44154,
+          "code_similarity": 0.175,
+          "size_similarity": 0.566,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.414,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44072,
+          "right_pr_number": 44722,
+          "code_similarity": 0.124,
+          "size_similarity": 0.357,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.347,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44085,
+          "right_pr_number": 44129,
+          "code_similarity": 0.141,
+          "size_similarity": 0.503,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.272,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44085,
+          "right_pr_number": 44154,
+          "code_similarity": 0.18,
+          "size_similarity": 0.684,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.289,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44085,
+          "right_pr_number": 44722,
+          "code_similarity": 0.728,
+          "size_similarity": 0.923,
+          "file_overlap": 0.5,
+          "area_overlap": 0.791,
+          "patch_similarity": 0.78,
+          "shared_filenames": [
+            "src/transformers/models/gptj/modeling_gptj.py"
+          ],
+          "shared_file_areas": [
+            {
+              "filename": "src/transformers/models/gptj/modeling_gptj.py",
+              "left_ranges": [
+                [
+                  33,
+                  40
+                ],
+                [
+                  173,
+                  180
+                ],
+                [
+                  249,
+                  256
+                ],
+                [
+                  397,
+                  405
+                ],
+                [
+                  408,
+                  419
+                ],
+                [
+                  425,
+                  434
+                ],
+                [
+                  459,
+                  465
+                ],
+                [
+                  470,
+                  489
+                ],
+                [
+                  517,
+                  539
+                ],
+                [
+                  553,
+                  559
+                ],
+                [
+                  565,
+                  574
+                ],
+                [
+                  579,
+                  597
+                ],
+                [
+                  600,
+                  611
+                ],
+                [
+                  633,
+                  639
+                ],
+                [
+                  645,
+                  652
+                ],
+                [
+                  657,
+                  673
+                ],
+                [
+                  716,
+                  728
+                ],
+                [
+                  737,
+                  743
+                ],
+                [
+                  748,
+                  771
+                ],
+                [
+                  789,
+                  794
+                ]
+              ],
+              "right_ranges": [
+                [
+                  33,
+                  42
+                ],
+                [
+                  175,
+                  182
+                ],
+                [
+                  251,
+                  258
+                ],
+                [
+                  399,
+                  420
+                ],
+                [
+                  426,
+                  435
+                ],
+                [
+                  460,
+                  467
+                ],
+                [
+                  472,
+                  489
+                ],
+                [
+                  519,
+                  540
+                ],
+                [
+                  554,
+                  560
+                ],
+                [
+                  566,
+                  575
+                ],
+                [
+                  580,
+                  598
+                ],
+                [
+                  601,
+                  612
+                ],
+                [
+                  634,
+                  640
+                ],
+                [
+                  646,
+                  653
+                ],
+                [
+                  658,
+                  663
+                ],
+                [
+                  666,
+                  674
+                ],
+                [
+                  717,
+                  722
+                ],
+                [
+                  738,
+                  744
+                ],
+                [
+                  749,
+                  772
+                ],
+                [
+                  790,
+                  795
+                ]
+              ]
+            }
+          ]
+        },
+        {
+          "left_pr_number": 44129,
+          "right_pr_number": 44154,
+          "code_similarity": 0.099,
+          "size_similarity": 0.344,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.199,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44129,
+          "right_pr_number": 44722,
+          "code_similarity": 0.146,
+          "size_similarity": 0.545,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.247,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        },
+        {
+          "left_pr_number": 44154,
+          "right_pr_number": 44722,
+          "code_similarity": 0.164,
+          "size_similarity": 0.631,
+          "file_overlap": 0.0,
+          "area_overlap": 0.0,
+          "patch_similarity": 0.255,
+          "shared_filenames": [],
+          "shared_file_areas": []
+        }
+      ]
+    },
+    {
+      "cluster_id": "cluster-41211-3",
+      "summary": "Cluster of 1 issues and 2 PRs centered on issue #41211.",
+      "status": "open",
+      "confidence": 0.55,
+      "canonical_issue_number": 41211,
+      "canonical_pr_number": 44339,
+      "issue_numbers": [
+        41211
+      ],
+      "pr_numbers": [
+        41356,
+        44339
+      ],
+      "evidence_types": [
+        "closing_reference"
+      ],
+      "pr_comparisons": [
+        {
+          "left_pr_number": 41356,
+          "right_pr_number": 44339,
+          "code_similarity": 0.155,
+          "size_similarity": 0.094,
+          "file_overlap": 0.273,
+          "area_overlap": 0.078,
+          "patch_similarity": 0.183,
+          "shared_filenames": [
+            "docs/source/en/model_doc/deimv2.md",
+            "src/transformers/models/deimv2/__init__.py",
+            "src/transformers/models/deimv2/configuration_deimv2.py",
+            "src/transformers/models/deimv2/modeling_deimv2.py",
+            "tests/models/deimv2/__init__.py",
+            "tests/models/deimv2/test_modeling_deimv2.py"
+          ],
+          "shared_file_areas": [
+            {
+              "filename": "docs/source/en/model_doc/deimv2.md",
+              "left_ranges": [
+                [
+                  1,
+                  132
+                ]
+              ],
+              "right_ranges": [
+                [
+                  1,
+                  65
+                ]
+              ]
+            },
+            {
+              "filename": "src/transformers/models/deimv2/__init__.py",
+              "left_ranges": [
+                [
+                  1,
+                  15
+                ]
+              ],
+              "right_ranges": [
+                [
+                  1,
+                  29
+                ]
+              ]
+            },
+            {
+              "filename": "src/transformers/models/deimv2/configuration_deimv2.py",
+              "left_ranges": [
+                [
+                  1,
+                  74
+                ]
+              ],
+              "right_ranges": [
+                [
+                  1,
+                  266
+                ]
+              ]
+            },
+            {
+              "filename": "tests/models/deimv2/test_modeling_deimv2.py",
+              "left_ranges": [
+                [
+                  1,
+                  15
+                ]
+              ],
+              "right_ranges": [
+                [
+                  1,
+                  1734
+                ]
+              ]
+            }
+          ]
+        }
+      ]
+    },
+    {
+      "cluster_id": "cluster-43240-3",
+      "summary": "Cluster of 1 issues and 2 PRs centered on issue #43240.",
+      "status": "open",
+      "confidence": 0.75,
+      "canonical_issue_number": 43240,
+      "canonical_pr_number": 43251,
+      "issue_numbers": [
+        43240
+      ],
+      "pr_numbers": [
+        43251,
+        43254
+      ],
+      "evidence_types": [
+        "closing_reference",
+        "shared_issue_target"
+      ],
+      "pr_comparisons": [
+        {
+          "left_pr_number": 43251,
+          "right_pr_number": 43254,
+          "code_similarity": 0.794,
+          "size_similarity": 0.64,
+          "file_overlap": 1.0,
+          "area_overlap": 0.667,
+          "patch_similarity": 0.882,
+          "shared_filenames": [
+            "src/transformers/loss/loss_utils.py"
+          ],
+          "shared_file_areas": [
+            {
+              "filename": "src/transformers/loss/loss_utils.py",
+              "left_ranges": [
+                [
+                  30,
+                  50
+                ]
+              ],
+              "right_ranges": [
+                [
+                  30,
+                  43
+                ]
+              ]
+            }
+          ]
+        }
+      ]
+    },
+    {
+      "cluster_id": "cluster-41115-3",
+      "summary": "Cluster of 1 issues and 2 PRs centered on issue #41115.",
+      "status": "open",
+      "confidence": 0.55,
+      "canonical_issue_number": 41115,
+      "canonical_pr_number": 45613,
+      "issue_numbers": [
+        41115
+      ],
+      "pr_numbers": [
+        41116,
+        45613
+      ],
+      "evidence_types": [
+        "closing_reference"
+      ],
+      "pr_comparisons": [
+        {
+          "left_pr_number": 41116,
+          "right_pr_number": 45613,
+          "code_similarity": 0.805,
+          "size_similarity": 0.885,
+          "file_overlap": 0.75,
+          "area_overlap": 0.854,
+          "patch_similarity": 0.693,
+          "shared_filenames": [
+            "docs/source/en/_toctree.yml",
+            "docs/source/en/model_doc/minicpm3.md",
+            "src/transformers/models/auto/modeling_auto.py",
+            "src/transformers/models/minicpm3/__init__.py",
+            "src/transformers/models/minicpm3/configuration_minicpm3.py",
+            "src/transformers/models/minicpm3/modeling_minicpm3.py",
+            "src/transformers/models/minicpm3/modular_minicpm3.py",
+            "tests/models/minicpm3/__init__.py",
+            "tests/models/minicpm3/test_modeling_minicpm3.py"
+          ],
+          "shared_file_areas": [
+            {
+              "filename": "docs/source/en/_toctree.yml",
+              "left_ranges": [
+                [
+                  713,
+                  720
+                ]
+              ],
+              "right_ranges": [
+                [
+                  713,
+                  720
+                ]
+              ]
+            },
+            {
+              "filename": "docs/source/en/model_doc/minicpm3.md",
+              "left_ranges": [
+                [
+                  1,
+                  67
+                ]
+              ],
+              "right_ranges": [
+                [
+                  1,
+                  45
+                ]
+              ]
+            },
+            {
+              "filename": "src/transformers/models/auto/modeling_auto.py",
+              "left_ranges": [
+                [
+                  285,
+                  291
+                ],
+                [
+                  700,
+                  706
+                ],
+                [
+                  1298,
+                  1304
+                ]
+              ],
+              "right_ranges": [
+                [
+                  699,
+                  705
+                ],
+                [
+                  1301,
+                  1307
+                ]
+              ]
+            },
+            {
+              "filename": "src/transformers/models/minicpm3/__init__.py",
+              "left_ranges": [
+                [
+                  1,
+                  29
+                ]
+              ],
+              "right_ranges": [
+                [
+                  1,
+                  29
+                ]
+              ]
+            },
+            {
+              "filename": "src/transformers/models/minicpm3/configuration_minicpm3.py",
+              "left_ranges": [
+                [
+                  1,
+                  141
+                ]
+              ],
+              "right_ranges": [
+                [
+                  1,
+                  126
+                ]
+              ]
+            },
+            {
+              "filename": "src/transformers/models/minicpm3/modeling_minicpm3.py",
+              "left_ranges": [
+                [
+                  1,
+                  544
+                ]
+              ],
+              "right_ranges": [
+                [
+                  1,
+                  522
+                ]
+              ]
+            },
+            {
+              "filename": "src/transformers/models/minicpm3/modular_minicpm3.py",
+              "left_ranges": [
+                [
+                  1,
+                  444
+                ]
+              ],
+              "right_ranges": [
+                [
+                  1,
+                  342
+                ]
+              ]
+            },
+            {
+              "filename": "tests/models/minicpm3/test_modeling_minicpm3.py",
+              "left_ranges": [
+                [
+                  1,
+                  109
+                ]
+              ],
+              "right_ranges": [
+                [
+                  1,
+                  136
+                ]
+              ]
+            }
+          ]
+        }
+      ]
+    },
+    {
+      "cluster_id": "cluster-45647-3",
+      "summary": "Cluster of 1 issues and 2 PRs centered on issue #45647.",
+      "status": "open",
+      "confidence": 0.75,
+      "canonical_issue_number": 45647,
+      "canonical_pr_number": 45738,
+      "issue_numbers": [
+        45647
+      ],
+      "pr_numbers": [
+        45737,
+        45738
+      ],
+      "evidence_types": [
+        "closing_reference",
+        "shared_issue_target"
+      ],
+      "pr_comparisons": [
+        {
+          "left_pr_number": 45737,
+          "right_pr_number": 45738,
+          "code_similarity": 0.526,
+          "size_similarity": 0.11,
+          "file_overlap": 0.5,
+          "area_overlap": 1.0,
+          "patch_similarity": 0.024,
+          "shared_filenames": [
+            "src/transformers/models/musicgen_melody/modeling_musicgen_melody.py"
+          ],
+          "shared_file_areas": [
+            {
+              "filename": "src/transformers/models/musicgen_melody/modeling_musicgen_melody.py",
+              "left_ranges": [
+                [
+                  491,
+                  497
+                ]
+              ],
+              "right_ranges": [
+                [
+                  491,
+                  497
+                ]
+              ]
+            }
+          ]
+        }
+      ]
+    },
+    {
+      "cluster_id": "cluster-43824-3",
+      "summary": "Cluster of 1 issues and 2 PRs centered on issue #43824.",
+      "status": "open",
+      "confidence": 0.75,
+      "canonical_issue_number": 43824,
+      "canonical_pr_number": 43836,
+      "issue_numbers": [
+        43824
+      ],
+      "pr_numbers": [
+        43836,
+        43842
+      ],
+      "evidence_types": [
+        "closing_reference",
+        "shared_issue_target"
+      ],
+      "pr_comparisons": [
+        {
+          "left_pr_number": 43836,
+          "right_pr_number": 43842,
+          "code_similarity": 0.332,
+          "size_similarity": 0.088,
+          "file_overlap": 1.0,
+          "area_overlap": 0.033,
+          "patch_similarity": 0.017,
+          "shared_filenames": [
+            "src/transformers/cli/serve.py"
+          ],
+          "shared_file_areas": [
+            {
+              "filename": "src/transformers/cli/serve.py",
+              "left_ranges": [
+                [
+                  11,
+                  18
+                ],
+                [
+                  359,
+                  429
+                ],
+                [
+                  584,
+                  590
+                ],
+                [
+                  1892,
+                  1910
+                ],
+                [
+                  1917,
+                  1923
+                ]
+              ],
+              "right_ranges": [
+                [
+                  54,
+                  61
+                ],
+                [
+                  587,
+                  593
+                ]
+              ]
+            }
+          ]
+        }
+      ]
+    },
+    {
+      "cluster_id": "cluster-45081-3",
+      "summary": "Cluster of 1 issues and 2 PRs centered on issue #45081.",
+      "status": "open",
+      "confidence": 0.8,
+      "canonical_issue_number": 45081,
+      "canonical_pr_number": 45317,
+      "issue_numbers": [
+        45081
+      ],
+      "pr_numbers": [
+        45086,
+        45317
+      ],
+      "evidence_types": [
+        "closing_reference",
+        "shared_issue_target",
+        "soft_similarity"
+      ],
+      "pr_comparisons": [
+        {
+          "left_pr_number": 45086,
+          "right_pr_number": 45317,
+          "code_similarity": 0.589,
+          "size_similarity": 0.34,
+          "file_overlap": 0.5,
+          "area_overlap": 1.0,
+          "patch_similarity": 0.136,
+          "shared_filenames": [
+            "src/transformers/tokenization_utils_tokenizers.py"
+          ],
+          "shared_file_areas": [
+            {
+              "filename": "src/transformers/tokenization_utils_tokenizers.py",
+              "left_ranges": [
+                [
+                  1360,
+                  1370
+                ],
+                [
+                  1374,
+                  1380
+                ]
+              ],
+              "right_ranges": [
+                [
+                  1360,
+                  1370
+                ],
+                [
+                  1374,
+                  1380
+                ]
+              ]
+            }
+          ]
+        }
+      ]
+    },
+    {
+      "cluster_id": "cluster-45561-3",
+      "summary": "Cluster of 1 issues and 2 PRs centered on issue #45561.",
+      "status": "open",
+      "confidence": 0.8,
+      "canonical_issue_number": 45561,
+      "canonical_pr_number": 45645,
+      "issue_numbers": [
+        45561
+      ],
+      "pr_numbers": [
+        45639,
+        45645
+      ],
+      "evidence_types": [
+        "closing_reference",
+        "shared_issue_target",
+        "soft_similarity"
+      ],
+      "pr_comparisons": [
+        {
+          "left_pr_number": 45639,
+          "right_pr_number": 45645,
+          "code_similarity": 0.532,
+          "size_similarity": 0.619,
+          "file_overlap": 0.5,
+          "area_overlap": 0.543,
+          "patch_similarity": 0.451,
+          "shared_filenames": [
+            "src/transformers/testing_utils.py",
+            "tests/utils/test_testing_utils.py"
+          ],
+          "shared_file_areas": [
+            {
+              "filename": "src/transformers/testing_utils.py",
+              "left_ranges": [
+                [
+                  3529,
+                  3536
+                ],
+                [
+                  3753,
+                  3787
+                ]
+              ],
+              "right_ranges": [
+                [
+                  3525,
+                  3558
+                ],
+                [
+                  3782,
+                  3788
+                ]
+              ]
+            },
+            {
+              "filename": "tests/utils/test_testing_utils.py",
+              "left_ranges": [
+                [
+                  1,
+                  86
+                ]
+              ],
+              "right_ranges": [
+                [
+                  1,
+                  114
+                ]
+              ]
+            }
+          ]
+        }
+      ]
+    },
+    {
+      "cluster_id": "cluster-43698-3",
+      "summary": "Cluster of 1 issues and 2 PRs centered on issue #43698.",
+      "status": "open",
+      "confidence": 0.75,
+      "canonical_issue_number": 43698,
+      "canonical_pr_number": 43779,
+      "issue_numbers": [
+        43698
+      ],
+      "pr_numbers": [
+        43779,
+        43816
+      ],
+      "evidence_types": [
+        "closing_reference",
+        "shared_issue_target"
+      ],
+      "pr_comparisons": [
+        {
+          "left_pr_number": 43779,
+          "right_pr_number": 43816,
+          "code_similarity": 0.418,
+          "size_similarity": 0.538,
+          "file_overlap": 1.0,
+          "area_overlap": 0.02,
+          "patch_similarity": 0.02,
+          "shared_filenames": [
+            "src/transformers/integrations/integration_utils.py"
+          ],
+          "shared_file_areas": [
+            {
+              "filename": "src/transformers/integrations/integration_utils.py",
+              "left_ranges": [
+                [
+                  2227,
+                  2241
+                ],
+                [
+                  2303,
+                  2309
+                ]
+              ],
+              "right_ranges": [
+                [
+                  2278,
+                  2291
+                ],
+                [
+                  2309,
+                  2322
+                ]
+              ]
+            }
+          ]
+        }
+      ]
+    },
+    {
+      "cluster_id": "cluster-44018-2",
+      "summary": "Cluster of 2 related pull requests linked by soft_similarity.",
+      "status": "open",
+      "confidence": 0.5,
+      "canonical_issue_number": null,
+      "canonical_pr_number": 44068,
+      "issue_numbers": [],
+      "pr_numbers": [
+        44018,
+        44068
+      ],
+      "evidence_types": [
+        "soft_similarity"
+      ],
+      "pr_comparisons": [
+        {
+          "left_pr_number": 44018,
+          "right_pr_number": 44068,
+          "code_similarity": 0.766,
+          "size_similarity": 0.939,
+          "file_overlap": 1.0,
+          "area_overlap": 0.425,
+          "patch_similarity": 0.866,
+          "shared_filenames": [
+            "src/transformers/models/gpt_neo/modeling_gpt_neo.py",
+            "tests/models/gpt_neo/test_modeling_gpt_neo.py"
+          ],
+          "shared_file_areas": [
+            {
+              "filename": "src/transformers/models/gpt_neo/modeling_gpt_neo.py",
+              "left_ranges": [
+                [
+                  26,
+                  46
+                ],
+                [
+                  138,
+                  143
+                ],
+                [
+                  182,
+                  187
+                ],
+                [
+                  283,
+                  295
+                ],
+                [
+                  327,
+                  341
+                ],
+                [
+                  348,
+                  354
+                ],
+                [
+                  360,
+                  369
+                ],
+                [
+                  399,
+                  406
+                ],
+                [
+                  411,
+                  419
+                ],
+                [
+                  428,
+                  433
+                ],
+                [
+                  472,
+                  492
+                ],
+                [
+                  507,
+                  513
+                ],
+                [
+                  519,
+                  528
+                ],
+                [
+                  541,
+                  559
+                ],
+                [
+                  562,
+                  567
+                ],
+                [
+                  595,
+                  601
+                ],
+                [
+                  607,
+                  614
+                ],
+                [
+                  627,
+                  643
+                ],
+                [
+                  685,
+                  690
+                ],
+                [
+                  708,
+                  714
+                ],
+                [
+                  720,
+                  727
+                ],
+                [
+                  740,
+                  757
+                ],
+                [
+                  761,
+                  766
+                ],
+                [
+                  780,
+                  786
+                ],
+                [
+                  791,
+                  798
+                ],
+                [
+                  807,
+                  822
+                ],
+                [
+                  840,
+                  845
+                ]
+              ],
+              "right_ranges": [
+                [
+                  26,
+                  31
+                ],
+                [
+                  34,
+                  44
+                ],
+                [
+                  136,
+                  141
+                ],
+                [
+                  180,
+                  185
+                ],
+                [
+                  281,
+                  293
+                ],
+                [
+                  325,
+                  339
+                ],
+                [
+                  346,
+                  352
+                ],
+                [
+                  358,
+                  367
+                ],
+                [
+                  397,
+                  404
+                ],
+                [
+                  409,
+                  417
+                ],
+                [
+                  426,
+                  434
+                ],
+                [
+                  463,
+                  483
+                ],
+                [
+                  498,
+                  504
+                ],
+                [
+                  510,
+                  519
+                ],
+                [
+                  532,
+                  550
+                ],
+                [
+                  553,
+                  564
+                ],
+                [
+                  586,
+                  592
+                ],
+                [
+                  598,
+                  605
+                ],
+                [
+                  618,
+                  634
+                ],
+                [
+                  676,
+                  688
+                ],
+                [
+                  699,
+                  705
+                ],
+                [
+                  711,
+                  718
+                ],
+                [
+                  731,
+                  748
+                ],
+                [
+                  752,
+                  762
+                ],
+                [
+                  771,
+                  777
+                ],
+                [
+                  782,
+                  789
+                ],
+                [
+                  798,
+                  813
+                ],
+                [
+                  831,
+                  836
+                ]
+              ]
+            },
+            {
+              "filename": "tests/models/gpt_neo/test_modeling_gpt_neo.py",
+              "left_ranges": [
+                [
+                  458,
+                  464
+                ]
+              ],
+              "right_ranges": [
+                [
+                  458,
+                  464
+                ]
+              ]
+            }
+          ]
+        }
+      ]
+    }
+  ],
+  "duplicate_issues": [],
+  "duplicate_prs": [
+    {
+      "cluster_id": "cluster-41115-3",
+      "canonical_pr_number": 45613,
+      "duplicate_pr_numbers": [
+        41116
+      ],
+      "target_issue_number": 41115,
+      "reason": "PRs in cluster-41115-3 are treated as duplicates because they converge on issue #41115 with closing_reference evidence."
+    },
+    {
+      "cluster_id": "cluster-41211-3",
+      "canonical_pr_number": 44339,
+      "duplicate_pr_numbers": [
+        41356
+      ],
+      "target_issue_number": 41211,
+      "reason": "PRs in cluster-41211-3 are treated as duplicates because they converge on issue #41211 with closing_reference evidence."
+    },
+    {
+      "cluster_id": "cluster-43240-3",
+      "canonical_pr_number": 43251,
+      "duplicate_pr_numbers": [
+        43254
+      ],
+      "target_issue_number": 43240,
+      "reason": "PRs in cluster-43240-3 are treated as duplicates because they converge on issue #43240 with closing_reference, shared_issue_target evidence."
+    },
+    {
+      "cluster_id": "cluster-43698-3",
+      "canonical_pr_number": 43779,
+      "duplicate_pr_numbers": [
+        43816
+      ],
+      "target_issue_number": 43698,
+      "reason": "PRs in cluster-43698-3 are treated as duplicates because they converge on issue #43698 with closing_reference, shared_issue_target evidence."
+    },
+    {
+      "cluster_id": "cluster-43824-3",
+      "canonical_pr_number": 43836,
+      "duplicate_pr_numbers": [
+        43842
+      ],
+      "target_issue_number": 43824,
+      "reason": "PRs in cluster-43824-3 are treated as duplicates because they converge on issue #43824 with closing_reference, shared_issue_target evidence."
+    },
+    {
+      "cluster_id": "cluster-43979-11",
+      "canonical_pr_number": 44007,
+      "duplicate_pr_numbers": [
+        43996,
+        44013,
+        44044,
+        44066,
+        44072,
+        44085,
+        44129,
+        44154,
+        44722
+      ],
+      "target_issue_number": 43979,
+      "reason": "PRs in cluster-43979-11 are treated as duplicates because they converge on issue #43979 with closing_reference, shared_issue_target, soft_similarity evidence."
+    },
+    {
+      "cluster_id": "cluster-44018-2",
+      "canonical_pr_number": 44068,
+      "duplicate_pr_numbers": [
+        44018
+      ],
+      "target_issue_number": null,
+      "reason": "PRs in cluster-44018-2 are treated as duplicates because they share soft_similarity evidence."
+    },
+    {
+      "cluster_id": "cluster-45081-3",
+      "canonical_pr_number": 45317,
+      "duplicate_pr_numbers": [
+        45086
+      ],
+      "target_issue_number": 45081,
+      "reason": "PRs in cluster-45081-3 are treated as duplicates because they converge on issue #45081 with closing_reference, shared_issue_target, soft_similarity evidence."
+    },
+    {
+      "cluster_id": "cluster-45561-3",
+      "canonical_pr_number": 45645,
+      "duplicate_pr_numbers": [
+        45639
+      ],
+      "target_issue_number": 45561,
+      "reason": "PRs in cluster-45561-3 are treated as duplicates because they converge on issue #45561 with closing_reference, shared_issue_target, soft_similarity evidence."
+    },
+    {
+      "cluster_id": "cluster-45647-3",
+      "canonical_pr_number": 45738,
+      "duplicate_pr_numbers": [
+        45737
+      ],
+      "target_issue_number": 45647,
+      "reason": "PRs in cluster-45647-3 are treated as duplicates because they converge on issue #45647 with closing_reference, shared_issue_target evidence."
+    }
+  ],
+  "best_issue": {
+    "cluster_id": "cluster-45647-3",
+    "issue_number": 45647,
+    "reason": "Issue #45647 is the strongest global issue candidate because it is open, belongs to a cluster with 3 artifacts, and carries 12 discussion comments plus 8 inbound references.",
+    "score": 162.0
+  },
+  "best_pr": {
+    "cluster_id": "cluster-43979-11",
+    "pr_number": 44007,
+    "reason": "PR #44007 is the strongest global PR candidate because it is open, belongs to a cluster with 11 artifacts, links to 1 issues, and carries 2 review/discussion events.",
+    "score": 238.886
+  }
+}
diff --git a/snapshots/20260510T060019Z/analysis-runs/hybrid-model-20260510t060019z/analysis-report-hybrid.llm-reviews.json b/snapshots/20260510T060019Z/analysis-runs/hybrid-model-20260510t060019z/analysis-report-hybrid.llm-reviews.json
new file mode 100644
index 0000000000000000000000000000000000000000..95c28fd896e0d289d7282bd98dee50b12a8147e2
--- /dev/null
+++ b/snapshots/20260510T060019Z/analysis-runs/hybrid-model-20260510t060019z/analysis-report-hybrid.llm-reviews.json
@@ -0,0 +1,19041 @@
+{
+  "schema_version": "1.0",
+  "repo": "huggingface/transformers",
+  "snapshot_id": "20260510T060019Z",
+  "generated_at": "2026-05-10T06:03:17Z",
+  "model": "gpt-5.4-mini?service_tier=flex",
+  "reviews": [
+    {
+      "label": "issue",
+      "component_index": 1,
+      "component_count": 16,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-13244-3",
+      "nodes": [
+        "issue:13244",
+        "issue:15980",
+        "issue:16618"
+      ],
+      "soft_pairs": [
+        "issue:13244|issue:16618",
+        "issue:15980|issue:16618"
+      ],
+      "prepared_review_unit_hash": "6ecb5e0f3a8a83d0a456d71df09f36534fc6fc3f219e64fd95715f0bbf3d4045",
+      "component_budget": {
+        "node_count": 3,
+        "item_count": 3,
+        "soft_pair_count": 2,
+        "serialized_chars": 2477,
+        "estimated_input_tokens": 620,
+        "estimated_eval_tokens": 1496
+      },
+      "budget": {
+        "node_count": 3,
+        "item_count": 3,
+        "soft_pair_count": 2,
+        "serialized_chars": 2477,
+        "estimated_input_tokens": 620,
+        "estimated_eval_tokens": 1496
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "These three issues are about different failure modes: Tapas tokenization behavior, a private-model download authentication/error-message problem, and a model-loading failure for bert-base-uncased. They do not look like the same underlying bug cluster.",
+        "confidence": 0.92,
+        "canonical_issue_reason": "Issue 16618 is the most active and most general user-facing report in the set, but it is still a different problem from the other two.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "16618 is the strongest representative only by discussion/activity, not because it matches the other issues.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:13244",
+            "right": "issue:16618",
+            "accept": false,
+            "reason": "Tapas tokenization mismatch is a different code path and symptom than a bert-base-uncased model-loading failure."
+          },
+          {
+            "left": "issue:15980",
+            "right": "issue:16618",
+            "accept": false,
+            "reason": "Private-model authentication/error-message handling is distinct from a generic inability to load bert-base-uncased."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 1,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:24643",
+        "issue:30277",
+        "issue:34634",
+        "issue:34928",
+        "issue:35545",
+        "issue:36010",
+        "issue:36331",
+        "issue:39290",
+        "issue:41093",
+        "issue:41720",
+        "issue:41762",
+        "issue:42915",
+        "issue:43039",
+        "issue:43404",
+        "issue:43425",
+        "issue:43531",
+        "issue:43541",
+        "issue:43596",
+        "issue:43638",
+        "issue:43716",
+        "issue:43792",
+        "issue:43827",
+        "issue:43828",
+        "issue:43854",
+        "issue:43866",
+        "issue:43901",
+        "issue:43975",
+        "issue:44292",
+        "issue:44322",
+        "issue:44457",
+        "issue:44509",
+        "issue:44512",
+        "issue:44560",
+        "issue:44805",
+        "issue:44841",
+        "issue:44863",
+        "issue:44918",
+        "issue:45070",
+        "issue:45081",
+        "issue:45084",
+        "issue:45161",
+        "issue:45237",
+        "issue:45362",
+        "issue:45464",
+        "issue:45507",
+        "issue:45685",
+        "issue:45800",
+        "issue:45823"
+      ],
+      "soft_pairs": [
+        "issue:44322|issue:45464",
+        "issue:45237|issue:45685",
+        "issue:43716|issue:45237",
+        "issue:43828|issue:45237",
+        "issue:45237|issue:45507",
+        "issue:44292|issue:45237",
+        "issue:30277|issue:35545",
+        "issue:35545|issue:42915",
+        "issue:35545|issue:41720",
+        "issue:44509|issue:44512",
+        "issue:35545|issue:36010",
+        "issue:44805|issue:45823",
+        "issue:43638|issue:44805",
+        "issue:43596|issue:44805",
+        "issue:44805|issue:45161",
+        "issue:44805|issue:44918",
+        "issue:34634|issue:35545",
+        "issue:30277|issue:43638",
+        "issue:43596|issue:43638",
+        "issue:43039|issue:45161",
+        "issue:43975|issue:44457",
+        "issue:44292|issue:44457",
+        "issue:34928|issue:44805",
+        "issue:24643|issue:30277",
+        "issue:43638|issue:45161",
+        "issue:34928|issue:36331",
+        "issue:43596|issue:45161",
+        "issue:43716|issue:45507",
+        "issue:39290|issue:45081",
+        "issue:43828|issue:45161",
+        "issue:43425|issue:44292",
+        "issue:44918|issue:45161",
+        "issue:34928|issue:41720",
+        "issue:43716|issue:45685",
+        "issue:43541|issue:43716",
+        "issue:43828|issue:44292",
+        "issue:45507|issue:45685",
+        "issue:43716|issue:44292",
+        "issue:44292|issue:44918",
+        "issue:43039|issue:44918",
+        "issue:41762|issue:43638",
+        "issue:43827|issue:44512",
+        "issue:43716|issue:44560",
+        "issue:43828|issue:45685",
+        "issue:44560|issue:45685",
+        "issue:43039|issue:43828",
+        "issue:30277|issue:36331",
+        "issue:34928|issue:36010",
+        "issue:43039|issue:44805",
+        "issue:43901|issue:44512",
+        "issue:41762|issue:43596",
+        "issue:39290|issue:41720",
+        "issue:44841|issue:45823",
+        "issue:43039|issue:44560",
+        "issue:43866|issue:44863",
+        "issue:43854|issue:43866",
+        "issue:39290|issue:43531",
+        "issue:41093|issue:43638",
+        "issue:43425|issue:45070",
+        "issue:43596|issue:45800",
+        "issue:43792|issue:45070",
+        "issue:45362|issue:45464",
+        "issue:43404|issue:43866",
+        "issue:39290|issue:45823",
+        "issue:45464|issue:45685",
+        "issue:43541|issue:45685",
+        "issue:43541|issue:43828",
+        "issue:36010|issue:39290",
+        "issue:44841|issue:45084",
+        "issue:41093|issue:41762",
+        "issue:45800|issue:45823"
+      ],
+      "prepared_review_unit_hash": "616cd81c38e8480ed7d8f53468498fdf7fd59fd92209ddc88a0170377250f824",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 71,
+        "serialized_chars": 48875,
+        "estimated_input_tokens": 12219,
+        "estimated_eval_tokens": 24694
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "Most pairs are false positives from broad lexical similarity. The only clear duplicate is the ZeRO-3 Bert index-out-of-bounds report pair; the rest differ by model, backend, or failure mode.",
+        "confidence": 0.78,
+        "canonical_issue_reason": "issue:43596 is the cleanest representative of the real duplicate pair: it describes the ZeRO-3/Bert index-out-of-bounds failure clearly and can serve as the canonical target for issue:43638.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "issue:43596 is the best issue to anchor duplicate triage because it is specific, reproducible, and has a true near-duplicate in issue:43638.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:44322",
+            "right": "issue:45464",
+            "accept": false,
+            "reason": "Same project area (Qwen3.5) but different failures: config attribute error vs streaming chat/completions failure."
+          },
+          {
+            "left": "issue:45237",
+            "right": "issue:45685",
+            "accept": false,
+            "reason": "Different models and backends: GPT-OSS on AMD GPUs vs an MPS histogram/int error."
+          },
+          {
+            "left": "issue:43716",
+            "right": "issue:45237",
+            "accept": false,
+            "reason": "Different bugs: Mistral-3 dtype mismatch in preprocessing vs GPT-OSS not working on AMD GPUs."
+          },
+          {
+            "left": "issue:43828",
+            "right": "issue:45237",
+            "accept": false,
+            "reason": "Both are dtype/back-end themed, but they are unrelated model-specific failures with no shared code path."
+          },
+          {
+            "left": "issue:45237",
+            "right": "issue:45507",
+            "accept": false,
+            "reason": "Different model families and symptoms: AMD GPU failure vs invalid method call in GraniteMoEHybrid."
+          },
+          {
+            "left": "issue:44292",
+            "right": "issue:45237",
+            "accept": false,
+            "reason": "Unrelated failures: Qwen-3 NVFP4 runtime error vs GPT-OSS AMD GPU issue."
+          },
+          {
+            "left": "issue:30277",
+            "right": "issue:35545",
+            "accept": false,
+            "reason": "Different subsystems entirely: DeepSpeed collective mismatch vs ONNX export failure."
+          },
+          {
+            "left": "issue:35545",
+            "right": "issue:42915",
+            "accept": false,
+            "reason": "Both are model/runtime issues, but one is ModernBERT ONNX export and the other is Qwen3Moe FP8 loading."
+          },
+          {
+            "left": "issue:35545",
+            "right": "issue:41720",
+            "accept": false,
+            "reason": "Different concrete problems: ONNX export vs auto device mapping cuda assert."
+          },
+          {
+            "left": "issue:44509",
+            "right": "issue:44512",
+            "accept": false,
+            "reason": "Both are v5 docs cleanup items, but they reference different removed docs/features and are not the same underlying issue."
+          },
+          {
+            "left": "issue:35545",
+            "right": "issue:36010",
+            "accept": false,
+            "reason": "Different code paths: ONNX export vs missing GenerationMixin import."
+          },
+          {
+            "left": "issue:44805",
+            "right": "issue:45823",
+            "accept": false,
+            "reason": "Both involve tensor shape/device problems, but the reports are about different models and different failure modes."
+          },
+          {
+            "left": "issue:43638",
+            "right": "issue:44805",
+            "accept": false,
+            "reason": "Similar error shape, but one is ZeRO-3 Bert index-out-of-bounds and the other is a generic mask/index shape mismatch."
+          },
+          {
+            "left": "issue:43596",
+            "right": "issue:44805",
+            "accept": false,
+            "reason": "Not the same bug: ZeRO-3 Bert loading failure vs a separate mask shape mismatch."
+          },
+          {
+            "left": "issue:44805",
+            "right": "issue:45161",
+            "accept": false,
+            "reason": "Generic shape-mismatch IndexError vs a GPT-OSS tensor-parallel issue."
+          },
+          {
+            "left": "issue:44805",
+            "right": "issue:44918",
+            "accept": false,
+            "reason": "Different failure mechanisms: mask/index shape mismatch vs Qwen3.5 embedding unpacking with TRL SFT."
+          },
+          {
+            "left": "issue:34634",
+            "right": "issue:35545",
+            "accept": false,
+            "reason": "BarkProcessor voice_preset bug is unrelated to ModernBERT ONNX export."
+          },
+          {
+            "left": "issue:30277",
+            "right": "issue:43638",
+            "accept": false,
+            "reason": "Both involve distributed loading, but the specific failures and model paths are different."
+          },
+          {
+            "left": "issue:43596",
+            "right": "issue:43638",
+            "accept": true,
+            "reason": "Same underlying ZeRO-3/Bert index-out-of-bounds bug; the titles differ only in model specificity."
+          },
+          {
+            "left": "issue:43039",
+            "right": "issue:45161",
+            "accept": false,
+            "reason": "Liger Kernel cross-entropy routing issue vs GPT-OSS tensor-parallel failure; not the same bug."
+          },
+          {
+            "left": "issue:43975",
+            "right": "issue:44457",
+            "accept": false,
+            "reason": "DeepSeek detokenization regression and LoRA save/load inconsistency are unrelated."
+          },
+          {
+            "left": "issue:44292",
+            "right": "issue:44457",
+            "accept": false,
+            "reason": "Different model/runtime issues: Qwen-3 NVFP4 error vs LoRA merge/reload mismatch."
+          },
+          {
+            "left": "issue:34928",
+            "right": "issue:44805",
+            "accept": false,
+            "reason": "Both are shape-related, but activation-checkpointing/FSDP mismatch is a different code path from the mask-index error."
+          },
+          {
+            "left": "issue:24643",
+            "right": "issue:30277",
+            "accept": false,
+            "reason": "DeepSpeed training weight-shape error is unrelated to the Jamba collective mismatch."
+          },
+          {
+            "left": "issue:43638",
+            "right": "issue:45161",
+            "accept": false,
+            "reason": "ZeRO-3 Bert loading failure vs GPT-OSS tensor-parallel issue."
+          },
+          {
+            "left": "issue:34928",
+            "right": "issue:36331",
+            "accept": false,
+            "reason": "Activation-checkpointing tensor mismatch and compute_loss signature mismatch are unrelated."
+          },
+          {
+            "left": "issue:43716",
+            "right": "issue:45507",
+            "accept": false,
+            "reason": "Mistral-3 dtype mismatch vs GraniteMoEHybrid invalid method call."
+          },
+          {
+            "left": "issue:39290",
+            "right": "issue:45081",
+            "accept": false,
+            "reason": "Gemma3 vLLM config attribute error is unrelated to the Mistral tokenizer regex patch crash."
+          },
+          {
+            "left": "issue:43828",
+            "right": "issue:45161",
+            "accept": false,
+            "reason": "Autocast dtype mismatch in Phi-tiny-MoE is not the same as the GPT-OSS tensor-parallel bug."
+          },
+          {
+            "left": "issue:43425",
+            "right": "issue:44292",
+            "accept": false,
+            "reason": "Torch version incompatibility and a Qwen-3 NVFP4 runtime issue are different problems."
+          },
+          {
+            "left": "issue:44918",
+            "right": "issue:45161",
+            "accept": false,
+            "reason": "TRL embedding unpacking issue is unrelated to GPT-OSS MoE tensor-parallel failure."
+          },
+          {
+            "left": "issue:34928",
+            "right": "issue:41720",
+            "accept": false,
+            "reason": "FSDP checkpointing mismatch and Qwen3 auto-device-map cuda assert are different failures."
+          },
+          {
+            "left": "issue:43716",
+            "right": "issue:45685",
+            "accept": false,
+            "reason": "Mistral-3 preprocessing dtype mismatch vs MPS histogram/int limitation."
+          },
+          {
+            "left": "issue:43541",
+            "right": "issue:43716",
+            "accept": false,
+            "reason": "Different model paths and error types: torch dynamo grouped_mm tracing vs image-preprocessor dtype mismatch."
+          },
+          {
+            "left": "issue:43828",
+            "right": "issue:44292",
+            "accept": false,
+            "reason": "Different models and failures: autocast dtype mismatch vs Qwen-3 NVFP4 runtime error."
+          },
+          {
+            "left": "issue:43039",
+            "right": "issue:44918",
+            "accept": false,
+            "reason": "Liger Kernel cross-entropy call issue is unrelated to Qwen3.5 embedding unpacking."
+          },
+          {
+            "left": "issue:41762",
+            "right": "issue:43638",
+            "accept": false,
+            "reason": "Both mention ZeRO-3 index errors, but Gemma3 loading and Bert loading are different concrete bug reports."
+          },
+          {
+            "left": "issue:43827",
+            "right": "issue:44512",
+            "accept": false,
+            "reason": "Both are docs regressions, but they concern different removed items and are not duplicates."
+          },
+          {
+            "left": "issue:43716",
+            "right": "issue:44560",
+            "accept": false,
+            "reason": "Different bugs: dtype mismatch vs video StopIteration in Qwen3-vl-embedding."
+          },
+          {
+            "left": "issue:43828",
+            "right": "issue:45685",
+            "accept": false,
+            "reason": "Autocast dtype mismatch and MPS histogram/int error are unrelated."
+          },
+          {
+            "left": "issue:44560",
+            "right": "issue:45685",
+            "accept": false,
+            "reason": "Different models and failure modes: video StopIteration vs an MPS kernel limitation."
+          },
+          {
+            "left": "issue:43039",
+            "right": "issue:43828",
+            "accept": false,
+            "reason": "Liger Kernel cross-entropy routing issue is not the same as Phi-tiny-MoE autocast dtype mismatch."
+          },
+          {
+            "left": "issue:30277",
+            "right": "issue:36331",
+            "accept": false,
+            "reason": "Distributed collective mismatch and trainer compute_loss kwarg incompatibility are unrelated."
+          },
+          {
+            "left": "issue:34928",
+            "right": "issue:36010",
+            "accept": false,
+            "reason": "Activation checkpointing/FSDP mismatch is unrelated to the GenerationMixin import error."
+          },
+          {
+            "left": "issue:43039",
+            "right": "issue:44805",
+            "accept": false,
+            "reason": "Liger Kernel cross-entropy issue is unrelated to the generic mask/index shape mismatch."
+          },
+          {
+            "left": "issue:43901",
+            "right": "issue:44512",
+            "accept": false,
+            "reason": "Both are documentation issues, but they refer to different deprecated APIs and different docs pages."
+          },
+          {
+            "left": "issue:41762",
+            "right": "issue:43596",
+            "accept": false,
+            "reason": "Same broad ZeRO-3 loading area, but Gemma3 and Bert reports are not clearly the same underlying bug."
+          },
+          {
+            "left": "issue:39290",
+            "right": "issue:41720",
+            "accept": false,
+            "reason": "Gemma3 text-config attribute error and Qwen3 auto-device-map cuda assert are different model-specific bugs."
+          },
+          {
+            "left": "issue:44841",
+            "right": "issue:45084",
+            "accept": false,
+            "reason": "Processor failure for Voxtral is unrelated to the template-node compilation error."
+          },
+          {
+            "left": "issue:41093",
+            "right": "issue:43638",
+            "accept": false,
+            "reason": "Mask/index shape mismatch and ZeRO-3 Bert index-out-of-bounds are different failures."
+          },
+          {
+            "left": "issue:43425",
+            "right": "issue:45070",
+            "accept": false,
+            "reason": "Torch version incompatibility and pydantic PretrainedConfig field regression are different issues."
+          },
+          {
+            "left": "issue:43596",
+            "right": "issue:45800",
+            "accept": false,
+            "reason": "ZeRO-3 Bert loading bug is unrelated to the torch/transformers incompatibility report."
+          },
+          {
+            "left": "issue:43792",
+            "right": "issue:45070",
+            "accept": false,
+            "reason": "Whisper loading/runtime issue is unrelated to the pydantic model regression."
+          },
+          {
+            "left": "issue:45362",
+            "right": "issue:45464",
+            "accept": false,
+            "reason": "Both involve Qwen3.5 and chat usage, but one is a model crash in transformers chat and the other is a streaming API failure."
+          },
+          {
+            "left": "issue:43404",
+            "right": "issue:43866",
+            "accept": false,
+            "reason": "Mistral3 lm_head tying bug and Ovis2 checkpoint corruption are unrelated."
+          },
+          {
+            "left": "issue:43854",
+            "right": "issue:43866",
+            "accept": false,
+            "reason": "GLM-4.7-Flash unit-test loading failure is unrelated to the Ovis2 checkpoint corruption report."
+          },
+          {
+            "left": "issue:43866",
+            "right": "issue:44863",
+            "accept": false,
+            "reason": "Different model families and load failures: corrupted checkpoint vs NemotronH checkpoint incompatibility."
+          },
+          {
+            "left": "issue:45507",
+            "right": "issue:45685",
+            "accept": false,
+            "reason": "GraniteMoEHybrid invalid method call is unrelated to the MPS histogram/int kernel issue."
+          },
+          {
+            "left": "issue:43716",
+            "right": "issue:44292",
+            "accept": false,
+            "reason": "Mistral-3 dtype mismatch and Qwen-3 NVFP4 runtime error are different model-specific bugs."
+          },
+          {
+            "left": "issue:44292",
+            "right": "issue:44918",
+            "accept": false,
+            "reason": "Qwen-3 NVFP4 runtime error and Qwen3.5/TRL embedding unpacking are unrelated."
+          },
+          {
+            "left": "issue:43039",
+            "right": "issue:44560",
+            "accept": false,
+            "reason": "Liger Kernel cross-entropy routing issue is unrelated to the Qwen3-vl StopIteration bug."
+          },
+          {
+            "left": "issue:44841",
+            "right": "issue:45084",
+            "accept": false,
+            "reason": "Voxtral processor failure and template-node compilation failure are different code paths."
+          }
+        ]
+      },
+      "evaluator_result": {
+        "accept": true,
+        "feedback": "The summary is grounded in the packet and stays conservative. The 43596/43638 pair is well supported as a near-duplicate ZeRO-3 Bert index-out-of-bounds report, and the other soft-edge rejections avoid overclaiming despite lexical similarity."
+      },
+      "evaluator_used": true,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": true,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 2,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:39186",
+        "issue:39692",
+        "issue:40990",
+        "issue:41093",
+        "issue:42915",
+        "issue:43054",
+        "issue:43296",
+        "issue:43366",
+        "issue:43404",
+        "issue:43531",
+        "issue:43541",
+        "issue:43572",
+        "issue:43596",
+        "issue:43644",
+        "issue:43645",
+        "issue:43716",
+        "issue:43828",
+        "issue:43866",
+        "issue:43873",
+        "issue:43950",
+        "issue:43975",
+        "issue:43994",
+        "issue:44291",
+        "issue:44292",
+        "issue:44360",
+        "issue:44387",
+        "issue:44403",
+        "issue:44492",
+        "issue:44512",
+        "issue:44530",
+        "issue:44534",
+        "issue:44589",
+        "issue:44863",
+        "issue:44918",
+        "issue:44928",
+        "issue:44960",
+        "issue:45005",
+        "issue:45020",
+        "issue:45070",
+        "issue:45161",
+        "issue:45399",
+        "issue:45464",
+        "issue:45507",
+        "issue:45542",
+        "issue:45632",
+        "issue:45685",
+        "issue:45698",
+        "issue:45800"
+      ],
+      "soft_pairs": [
+        "issue:43296|issue:43366",
+        "issue:43572|issue:45070",
+        "issue:44387|issue:45005",
+        "issue:43541|issue:45161",
+        "issue:43828|issue:45464",
+        "issue:44291|issue:45070",
+        "issue:43366|issue:45464",
+        "issue:45020|issue:45800",
+        "issue:43531|issue:45070",
+        "issue:43541|issue:44589",
+        "issue:41093|issue:43596",
+        "issue:43054|issue:43994",
+        "issue:45542|issue:45800",
+        "issue:43716|issue:44918",
+        "issue:44918|issue:45685",
+        "issue:45632|issue:45698",
+        "issue:43404|issue:45005",
+        "issue:43645|issue:45399",
+        "issue:40990|issue:44960",
+        "issue:44530|issue:45005",
+        "issue:44360|issue:44512",
+        "issue:42915|issue:45399",
+        "issue:44863|issue:45005",
+        "issue:43873|issue:45005",
+        "issue:39186|issue:44918",
+        "issue:44492|issue:44512",
+        "issue:44918|issue:45507",
+        "issue:43975|issue:44292",
+        "issue:43950|issue:44534",
+        "issue:43644|issue:43950",
+        "issue:44292|issue:45507",
+        "issue:39692|issue:43054",
+        "issue:43828|issue:44928",
+        "issue:43866|issue:44403"
+      ],
+      "prepared_review_unit_hash": "56e0ea3808b14e624c576e290464fb05ccb11227a14cb973ef75bd093dd0c803",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 34,
+        "serialized_chars": 37475,
+        "estimated_input_tokens": 9369,
+        "estimated_eval_tokens": 18994
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "Most items are distinct model- or release-specific bugs. The only clear duplicate set is the v5 non-persistent-buffer regression trio.",
+        "confidence": 0.84,
+        "canonical_issue_reason": "Issue 43950 is the clearest, most specific statement of the shared bug: `from_pretrained()` corrupts non-persistent buffers in Transformers 5.x.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "Issue 43950 is the best representative issue because it names the concrete failing path (`from_pretrained`) and the regression symptom directly.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:43296",
+            "right": "issue:43366",
+            "accept": false,
+            "reason": "Different failures: PaddleOCR-VL loading in vLLM vs gpt-oss GGUF architecture support."
+          },
+          {
+            "left": "issue:43572",
+            "right": "issue:45070",
+            "accept": false,
+            "reason": "Both are v5 config regressions, but the missing `pad_token_idx` and the pydantic `PretrainedConfig` breakage are different bugs."
+          },
+          {
+            "left": "issue:44387",
+            "right": "issue:45005",
+            "accept": false,
+            "reason": "Different problem spaces: int4 memory/OOM vs tied weights in translation models."
+          },
+          {
+            "left": "issue:43541",
+            "right": "issue:45161",
+            "accept": false,
+            "reason": "Both involve MoE models, but one is a grouped_mm tracing error and the other is a tensor-parallel support issue."
+          },
+          {
+            "left": "issue:43828",
+            "right": "issue:45464",
+            "accept": false,
+            "reason": "Different code paths and symptoms: autocast dtype mismatch vs chat/completions streaming failure."
+          },
+          {
+            "left": "issue:44291",
+            "right": "issue:45070",
+            "accept": false,
+            "reason": "`init_empty_weights` argument handling is unrelated to the `PretrainedConfig` pydantic regression."
+          },
+          {
+            "left": "issue:43366",
+            "right": "issue:45464",
+            "accept": false,
+            "reason": "Model/architecture support vs streaming API failure; no shared concrete bug."
+          },
+          {
+            "left": "issue:45020",
+            "right": "issue:45800",
+            "accept": false,
+            "reason": "Generic `remote_code` breakage vs torch/transformers version incompatibility are not the same issue."
+          },
+          {
+            "left": "issue:43531",
+            "right": "issue:45070",
+            "accept": false,
+            "reason": "Qwen3-MoE sliding-window behavior is unrelated to the config-field regression."
+          },
+          {
+            "left": "issue:43541",
+            "right": "issue:44589",
+            "accept": false,
+            "reason": "Tracing/grouped_mm failure is unrelated to missing Float8 storage support."
+          },
+          {
+            "left": "issue:41093",
+            "right": "issue:43596",
+            "accept": false,
+            "reason": "Both are IndexErrors, but the mask-shape mismatch and zero-sized tensor indexing are different bugs."
+          },
+          {
+            "left": "issue:43054",
+            "right": "issue:43994",
+            "accept": false,
+            "reason": "Both mention SigLIP2, but one is about worse text embeddings and the other about nonsensical AutoModel/pipeline outputs; not clearly the same concrete defect."
+          },
+          {
+            "left": "issue:45542",
+            "right": "issue:45800",
+            "accept": false,
+            "reason": "TensorBoard-without-TensorFlow backend handling is unrelated to the torch/transformers incompatibility."
+          },
+          {
+            "left": "issue:43716",
+            "right": "issue:44918",
+            "accept": false,
+            "reason": "Different failures: image preprocessor dtype mismatch vs TRL SFT embedding unpacking."
+          },
+          {
+            "left": "issue:44918",
+            "right": "issue:45685",
+            "accept": false,
+            "reason": "Unpacking embeddings in Qwen3.5 and MPS histogram errors are unrelated MoE issues."
+          },
+          {
+            "left": "issue:45632",
+            "right": "issue:45698",
+            "accept": false,
+            "reason": "Both involve custom-code loading, but one is a leaf-directory cache collision and the other is wrong-module selection after save_pretrained; not the same concrete bug."
+          },
+          {
+            "left": "issue:43404",
+            "right": "issue:45005",
+            "accept": false,
+            "reason": "Both mention tied weights, but the model families and failure modes differ."
+          },
+          {
+            "left": "issue:43645",
+            "right": "issue:45399",
+            "accept": false,
+            "reason": "Jupyter custom-model initialization and flash-attn fallback logic are unrelated."
+          },
+          {
+            "left": "issue:40990",
+            "right": "issue:44960",
+            "accept": false,
+            "reason": "Unrelated model/perplexity report vs a GLM5 issue."
+          },
+          {
+            "left": "issue:44530",
+            "right": "issue:45005",
+            "accept": false,
+            "reason": "PagedAttentionCache linear_attention crash is unrelated to tied-weight translation issues."
+          },
+          {
+            "left": "issue:44360",
+            "right": "issue:44512",
+            "accept": false,
+            "reason": "An algorithmic DSA-indexer discussion is unrelated to a docs typo."
+          },
+          {
+            "left": "issue:42915",
+            "right": "issue:45399",
+            "accept": false,
+            "reason": "Qwen3-MoE FP8 failure is unrelated to flash-attn fallback checks."
+          },
+          {
+            "left": "issue:44863",
+            "right": "issue:45005",
+            "accept": false,
+            "reason": "NemotronH checkpoint loading and tied-weight translation regressions are different bugs."
+          },
+          {
+            "left": "issue:43873",
+            "right": "issue:45005",
+            "accept": false,
+            "reason": "Quantization/offloading behavior is unrelated to tied-weight handling."
+          },
+          {
+            "left": "issue:39186",
+            "right": "issue:44918",
+            "accept": false,
+            "reason": "FSDP 2-D weight error is unrelated to TRL embedding unpacking."
+          },
+          {
+            "left": "issue:44492",
+            "right": "issue:44512",
+            "accept": false,
+            "reason": "A cache-strategy typo is unrelated to the removed run-command docs issue."
+          },
+          {
+            "left": "issue:44918",
+            "right": "issue:45507",
+            "accept": false,
+            "reason": "Different failures in different model families; no shared code-path bug is evident."
+          },
+          {
+            "left": "issue:43975",
+            "right": "issue:44292",
+            "accept": false,
+            "reason": "Detokenization regression and `init_empty_weights` argument handling are unrelated."
+          },
+          {
+            "left": "issue:43950",
+            "right": "issue:44534",
+            "accept": true,
+            "reason": "Same non-persistent-buffer regression in Transformers v5: both say `from_pretrained`/v5 fills or corrupts non-persistent buffers with junk."
+          },
+          {
+            "left": "issue:43644",
+            "right": "issue:43950",
+            "accept": true,
+            "reason": "Same underlying bug: Transformers v5 overwrites/corrupts non-persistent buffers during model loading."
+          },
+          {
+            "left": "issue:44292",
+            "right": "issue:45507",
+            "accept": false,
+            "reason": "Unexpected `_is_hf_initialized` argument and invalid method calls on GraniteMoEHybrid are unrelated."
+          },
+          {
+            "left": "issue:39692",
+            "right": "issue:43054",
+            "accept": false,
+            "reason": "SigLIP2 docs-example errors and SigLIP2 embedding quality complaints are related only by model family, not the same concrete bug."
+          },
+          {
+            "left": "issue:43828",
+            "right": "issue:44928",
+            "accept": false,
+            "reason": "Both mention dtype/precision problems, but autocast mismatch and RLHF NaN from 3D position_ids are different bugs."
+          },
+          {
+            "left": "issue:43866",
+            "right": "issue:44403",
+            "accept": false,
+            "reason": "Checkpoint corruption and loading noise are not the same issue."
+          }
+        ]
+      },
+      "evaluator_result": {
+        "accept": true,
+        "feedback": "The summary is grounded in the packet, and the only accepted cluster is well supported by the titles: 43644, 43950, and 44534 all describe the same Transformers v5 non-persistent-buffer corruption/regression. The rejected soft-edge pairs are stated conservatively and do not overclaim duplication from the available evidence."
+      },
+      "evaluator_used": true,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": true,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 3,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:36331",
+        "issue:38175",
+        "issue:41553",
+        "issue:41628",
+        "issue:41762",
+        "issue:42491",
+        "issue:42915",
+        "issue:42947",
+        "issue:43054",
+        "issue:43404",
+        "issue:43493",
+        "issue:43638",
+        "issue:43643",
+        "issue:43742",
+        "issue:43782",
+        "issue:43792",
+        "issue:43824",
+        "issue:43827",
+        "issue:43828",
+        "issue:43856",
+        "issue:43872",
+        "issue:43881",
+        "issue:43883",
+        "issue:43950",
+        "issue:43975",
+        "issue:44220",
+        "issue:44368",
+        "issue:44387",
+        "issue:44403",
+        "issue:44451",
+        "issue:44488",
+        "issue:44509",
+        "issue:44589",
+        "issue:44661",
+        "issue:44910",
+        "issue:44960",
+        "issue:44991",
+        "issue:45020",
+        "issue:45081",
+        "issue:45237",
+        "issue:45356",
+        "issue:45399",
+        "issue:45520",
+        "issue:45685",
+        "issue:45701",
+        "issue:45800",
+        "issue:45823",
+        "issue:45859"
+      ],
+      "soft_pairs": [
+        "issue:36331|issue:41762",
+        "issue:41628|issue:45800",
+        "issue:41553|issue:43883",
+        "issue:43872|issue:44589",
+        "issue:38175|issue:43054",
+        "issue:43054|issue:45701",
+        "issue:44910|issue:45859",
+        "issue:41553|issue:43881",
+        "issue:42947|issue:44387",
+        "issue:44403|issue:44451",
+        "issue:43975|issue:45520",
+        "issue:43643|issue:45020",
+        "issue:43054|issue:43493",
+        "issue:43975|issue:45237",
+        "issue:44403|issue:44488",
+        "issue:42947|issue:43856",
+        "issue:43881|issue:43883",
+        "issue:41762|issue:42915",
+        "issue:42491|issue:44368",
+        "issue:45685|issue:45823",
+        "issue:43975|issue:44451",
+        "issue:43824|issue:43872",
+        "issue:45020|issue:45823",
+        "issue:44991|issue:45399",
+        "issue:44661|issue:45020",
+        "issue:43828|issue:45823",
+        "issue:45020|issue:45356",
+        "issue:43782|issue:44960",
+        "issue:45081|issue:45399",
+        "issue:43827|issue:44509",
+        "issue:43404|issue:43950",
+        "issue:43792|issue:44220",
+        "issue:43742|issue:45399",
+        "issue:42915|issue:45020",
+        "issue:43638|issue:45823"
+      ],
+      "prepared_review_unit_hash": "846505c88f5c08bd3a8de30cce1ae3850e2737da3adaec4e0e33fdbaf5a3ad51",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 35,
+        "serialized_chars": 38226,
+        "estimated_input_tokens": 9557,
+        "estimated_eval_tokens": 19370
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "This cluster is mostly a loose collection of unrelated Transformers regression reports. I only see one plausible duplicate-style link: the two SigLIP2 issues, which both point to an HF-vs-reference implementation mismatch. Most other soft pairs are same broad area but not the same concrete bug.",
+        "confidence": 0.58,
+        "canonical_issue_reason": "issue:43493 is the best root-cause report for the only convincing duplicate pair: it explicitly describes the SigLIP2 HF-vs-JAX implementation discrepancy, while issue:43054 reads more like a symptom report about degraded text embeddings.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "issue:45020 is the strongest cluster anchor overall because it is an open umbrella report for recent remote_code-related breakages and best matches the cluster\u2019s repeated loading/version-regression theme, even though many items are still not true duplicates.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:36331",
+            "right": "issue:41762",
+            "accept": false,
+            "reason": "Different failures: trainer API change vs DeepSpeed ZeRO-3 loading index error."
+          },
+          {
+            "left": "issue:41628",
+            "right": "issue:45800",
+            "accept": false,
+            "reason": "Import error for AutoImageProcessor is unrelated to torch/transformers compatibility."
+          },
+          {
+            "left": "issue:41553",
+            "right": "issue:43883",
+            "accept": false,
+            "reason": "Different models and code paths: Voxtral AutoTokenizer error vs Molmo tied-weights attribute error."
+          },
+          {
+            "left": "issue:43872",
+            "right": "issue:44589",
+            "accept": false,
+            "reason": "Both are incompatibility reports, but the concrete errors and subsystems differ."
+          },
+          {
+            "left": "issue:38175",
+            "right": "issue:43054",
+            "accept": false,
+            "reason": "Both involve SigLIP/SigLIP2, but one is zero probabilities and the other is general embedding quality; not clearly the same bug."
+          },
+          {
+            "left": "issue:43054",
+            "right": "issue:45701",
+            "accept": false,
+            "reason": "SigLIP2 embedding quality issue vs a generic tokenization regression report."
+          },
+          {
+            "left": "issue:44910",
+            "right": "issue:45859",
+            "accept": false,
+            "reason": "Different Qwen3.5 MoE problems: flash-attn packed-sequence handling vs CPU offload incompatibility."
+          },
+          {
+            "left": "issue:41553",
+            "right": "issue:43881",
+            "accept": false,
+            "reason": "Unrelated model-loading failures for different models."
+          },
+          {
+            "left": "issue:42947",
+            "right": "issue:44387",
+            "accept": false,
+            "reason": "Gradient checkpointing/LoRA issue is not the same as int4 CUDA memory growth."
+          },
+          {
+            "left": "issue:44403",
+            "right": "issue:44451",
+            "accept": false,
+            "reason": "General loading noise vs a specific model load failure; not the same underlying bug."
+          },
+          {
+            "left": "issue:43975",
+            "right": "issue:45520",
+            "accept": false,
+            "reason": "Tokenizer detokenization regression and flash_attn import KeyError are unrelated."
+          },
+          {
+            "left": "issue:43643",
+            "right": "issue:45020",
+            "accept": false,
+            "reason": "Both mention remote_code, but one is missing fields in AutoConfig while the other is a broad compatibility report."
+          },
+          {
+            "left": "issue:43054",
+            "right": "issue:43493",
+            "accept": true,
+            "reason": "Both point to the same SigLIP2 implementation discrepancy; one is the symptom, the other the root-cause report."
+          },
+          {
+            "left": "issue:43975",
+            "right": "issue:45237",
+            "accept": false,
+            "reason": "Deepseek tokenizer behavior vs GPT-OSS-20B on AMD GPUs are unrelated."
+          },
+          {
+            "left": "issue:44403",
+            "right": "issue:44488",
+            "accept": false,
+            "reason": "Both are load-failure-ish, but they involve different models and no shared concrete mechanism."
+          },
+          {
+            "left": "issue:42947",
+            "right": "issue:43856",
+            "accept": false,
+            "reason": "LoRA gradient checkpointing and MoE memory usage are different training issues."
+          },
+          {
+            "left": "issue:43881",
+            "right": "issue:43883",
+            "accept": false,
+            "reason": "Different model families and different failure modes."
+          },
+          {
+            "left": "issue:41762",
+            "right": "issue:42915",
+            "accept": false,
+            "reason": "Gemma3 ZeRO-3 load error and Qwen3 MoE FP8 failure are unrelated."
+          },
+          {
+            "left": "issue:42491",
+            "right": "issue:44368",
+            "accept": false,
+            "reason": "Different Qwen3/Qwen3.5 training warnings and failure modes; not the same bug."
+          },
+          {
+            "left": "issue:45685",
+            "right": "issue:45823",
+            "accept": false,
+            "reason": "MPS histogram/int dtype issue is unrelated to Gemma4 device mismatch."
+          },
+          {
+            "left": "issue:43975",
+            "right": "issue:44451",
+            "accept": false,
+            "reason": "Tokenizer detokenization regression vs model loading failure are not the same issue."
+          },
+          {
+            "left": "issue:43824",
+            "right": "issue:43872",
+            "accept": false,
+            "reason": "Import error for Qwen2.5-VL class is unrelated to bitsandbytes Int8Params incompatibility."
+          },
+          {
+            "left": "issue:45020",
+            "right": "issue:45823",
+            "accept": false,
+            "reason": "Broad remote_code regressions and Gemma4 device placement are different bugs."
+          },
+          {
+            "left": "issue:44991",
+            "right": "issue:45399",
+            "accept": false,
+            "reason": "Tokenizer loading failure vs flash-attn fallback gating are unrelated."
+          },
+          {
+            "left": "issue:44661",
+            "right": "issue:45020",
+            "accept": false,
+            "reason": "Tokenizer mapping-name failure is a different code path from remote_code breakages."
+          },
+          {
+            "left": "issue:43828",
+            "right": "issue:45823",
+            "accept": false,
+            "reason": "Autocast dtype mismatch in Phi MoE and Gemma4 PLE device mismatch are different problems."
+          },
+          {
+            "left": "issue:45020",
+            "right": "issue:45356",
+            "accept": false,
+            "reason": "Remote_code breakages and Kimi-K2.5 tokenizer regression are related only at a very broad version-regression level."
+          },
+          {
+            "left": "issue:43792",
+            "right": "issue:44220",
+            "accept": false,
+            "reason": "Whisper runtime failure and fbank feature extraction issue may be related, but not clearly the same underlying bug from titles alone."
+          },
+          {
+            "left": "issue:45081",
+            "right": "issue:45399",
+            "accept": false,
+            "reason": "Mistral regex patch crash and flash-attn fallback blocking are unrelated."
+          },
+          {
+            "left": "issue:43827",
+            "right": "issue:44509",
+            "accept": false,
+            "reason": "Both are docs-related pipeline references, but they are separate documentation cleanup reports."
+          },
+          {
+            "left": "issue:43404",
+            "right": "issue:43950",
+            "accept": false,
+            "reason": "lm_head tying bug and non-persistent buffer corruption are different regression types."
+          },
+          {
+            "left": "issue:43792",
+            "right": "issue:43872",
+            "accept": false,
+            "reason": "Whisper/fbank failure and bitsandbytes storage-type incompatibility are unrelated."
+          },
+          {
+            "left": "issue:43742",
+            "right": "issue:45399",
+            "accept": false,
+            "reason": "MobileLLM load error vs flash-attn fallback gating are different issues."
+          },
+          {
+            "left": "issue:42915",
+            "right": "issue:45020",
+            "accept": false,
+            "reason": "Qwen3 MoE FP8 failure and remote_code model-loading regressions are not the same bug."
+          },
+          {
+            "left": "issue:43638",
+            "right": "issue:45823",
+            "accept": false,
+            "reason": "ZeRO-3 Bert load failure and Gemma4 device mismatch are unrelated."
+          }
+        ]
+      },
+      "evaluator_result": {
+        "accept": true,
+        "feedback": "The summary is cautious and supported by the packet. The only accepted soft edge, issue:43054 \u2194 issue:43493, is plausibly the same SigLIP2 discrepancy, and the other pair verdicts stay conservative by not over-duplicating broad but distinct regression reports."
+      },
+      "evaluator_used": true,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": true,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 4,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:24643",
+        "issue:30064",
+        "issue:36010",
+        "issue:36331",
+        "issue:38175",
+        "issue:41093",
+        "issue:41553",
+        "issue:42617",
+        "issue:42915",
+        "issue:43257",
+        "issue:43366",
+        "issue:43531",
+        "issue:43550",
+        "issue:43632",
+        "issue:43645",
+        "issue:43646",
+        "issue:43824",
+        "issue:43828",
+        "issue:43854",
+        "issue:43866",
+        "issue:43872",
+        "issue:43940",
+        "issue:43950",
+        "issue:43957",
+        "issue:43994",
+        "issue:44291",
+        "issue:44479",
+        "issue:44488",
+        "issue:44530",
+        "issue:44560",
+        "issue:44589",
+        "issue:44661",
+        "issue:44805",
+        "issue:44863",
+        "issue:44910",
+        "issue:44918",
+        "issue:44928",
+        "issue:44991",
+        "issue:45161",
+        "issue:45305",
+        "issue:45313",
+        "issue:45325",
+        "issue:45538",
+        "issue:45588",
+        "issue:45663",
+        "issue:45823",
+        "issue:45834"
+      ],
+      "soft_pairs": [
+        "issue:43257|issue:43866",
+        "issue:43940|issue:45313",
+        "issue:43872|issue:44291",
+        "issue:45161|issue:45823",
+        "issue:41553|issue:44991",
+        "issue:42915|issue:43872",
+        "issue:43854|issue:44863",
+        "issue:44928|issue:45588",
+        "issue:36331|issue:41093",
+        "issue:43257|issue:45305",
+        "issue:41553|issue:44488",
+        "issue:43872|issue:45538",
+        "issue:44530|issue:44863",
+        "issue:41093|issue:44805",
+        "issue:44479|issue:44560",
+        "issue:24643|issue:30064",
+        "issue:43366|issue:43828",
+        "issue:45663|issue:45823",
+        "issue:43366|issue:43531",
+        "issue:43950|issue:45834",
+        "issue:43550|issue:44910",
+        "issue:43645|issue:43646",
+        "issue:43366|issue:44589",
+        "issue:43950|issue:45325",
+        "issue:44560|issue:44918",
+        "issue:43824|issue:43957",
+        "issue:42617|issue:43366",
+        "issue:42915|issue:43957",
+        "issue:36010|issue:42915",
+        "issue:44661|issue:45823",
+        "issue:38175|issue:43994",
+        "issue:43632|issue:45834"
+      ],
+      "prepared_review_unit_hash": "f37d05c5d27101ba0be54f7851f3f289cf4c68d912396e16e19eab803a045e2d",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 47,
+        "item_count": 47,
+        "soft_pair_count": 32,
+        "serialized_chars": 36775,
+        "estimated_input_tokens": 9194,
+        "estimated_eval_tokens": 18644
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "The bucket is mostly heterogeneous, with only a few true duplicate subgroups. Clear matches exist for custom model initialization regressions, the _is_hf_initialized / init_empty_weights regression family, the mask-shape mismatch pair, and a couple of Qwen3/Qwen3.5 DeepSpeed or video-input regressions. No single issue cleanly represents the entire cluster.",
+        "confidence": 0.77,
+        "canonical_issue_reason": "43646 is the best canonical anchor among the actual duplicate subgroup because it states the core custom-model initialization regression broadly; 43645 is the notebook-specific variant of the same bug.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "Issue 43646 is the strongest representative of a true duplicate pair in this set, but the overall cluster is not a single coherent duplicate group.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:43257",
+            "right": "issue:43866",
+            "accept": false,
+            "reason": "Different problems: Qwen3 MoE weight conversion under accelerate/deepspeed vs a corrupted Ovis2 checkpoint."
+          },
+          {
+            "left": "issue:43940",
+            "right": "issue:45313",
+            "accept": true,
+            "reason": "Both describe DeepSpeed ZeRO-3 failing to load Qwen3-family weights, i.e. the same loader-path bug."
+          },
+          {
+            "left": "issue:43872",
+            "right": "issue:44291",
+            "accept": true,
+            "reason": "Same _is_hf_initialized / init_empty_weights regression; 43872 is a bitsandbytes manifestation of the same bug."
+          },
+          {
+            "left": "issue:45161",
+            "right": "issue:45823",
+            "accept": false,
+            "reason": "Both involve model-parallel-ish issues, but the concrete failures differ: TP on GPT-OSS MoE vs Gemma4 PLE device placement."
+          },
+          {
+            "left": "issue:41553",
+            "right": "issue:44991",
+            "accept": false,
+            "reason": "Both are tokenizer-related, but one is a bad error message for Voxtral and the other is a tokenizer load failure for a different model."
+          },
+          {
+            "left": "issue:42915",
+            "right": "issue:43872",
+            "accept": false,
+            "reason": "Different failure modes and code paths: Qwen3Moe with FineGrainedFP8Config vs a global _is_hf_initialized/bitsandbytes incompatibility."
+          },
+          {
+            "left": "issue:43854",
+            "right": "issue:44863",
+            "accept": false,
+            "reason": "Both are model-loading complaints, but they target different architectures and likely different loader bugs."
+          },
+          {
+            "left": "issue:44928",
+            "right": "issue:45588",
+            "accept": false,
+            "reason": "Qwen3.5 RLHF/Nan instability from 3D position_ids is unrelated to a flash_attention AttributeError on sink-less models."
+          },
+          {
+            "left": "issue:36331",
+            "right": "issue:41093",
+            "accept": false,
+            "reason": "Unexpected trainer kwarg and a mask/index shape mismatch are different bugs."
+          },
+          {
+            "left": "issue:43257",
+            "right": "issue:45305",
+            "accept": false,
+            "reason": "Qwen3 MoE weight conversion and DeepSpeed gradient averaging are separate training issues."
+          },
+          {
+            "left": "issue:41553",
+            "right": "issue:44488",
+            "accept": false,
+            "reason": "These are different tokenizer-loading problems for different models; no shared concrete code-path is evident."
+          },
+          {
+            "left": "issue:43872",
+            "right": "issue:45538",
+            "accept": false,
+            "reason": "A _is_hf_initialized constructor regression is unrelated to CLIPTokenizer's model_max_length default."
+          },
+          {
+            "left": "issue:44530",
+            "right": "issue:44863",
+            "accept": false,
+            "reason": "PagedAttentionCache linear_attention crash and NemotronH checkpoint loading are different subsystems."
+          },
+          {
+            "left": "issue:41093",
+            "right": "issue:44805",
+            "accept": true,
+            "reason": "Same indexed-tensor vs mask-length mismatch; 44805 looks like a duplicate report of the same error."
+          },
+          {
+            "left": "issue:44479",
+            "right": "issue:44560",
+            "accept": true,
+            "reason": "Both are Qwen VL video-input regressions in transformers 5.3.0; 44560 is a specific stopiteration symptom of the broader regression."
+          },
+          {
+            "left": "issue:24643",
+            "right": "issue:30064",
+            "accept": false,
+            "reason": "DeepSpeed training weight-shape error vs void segmentation-map processing are unrelated."
+          },
+          {
+            "left": "issue:43366",
+            "right": "issue:43828",
+            "accept": false,
+            "reason": "GGUF/gpt-oss support and a Phi-tiny-MoE dtype mismatch are unrelated."
+          },
+          {
+            "left": "issue:45663",
+            "right": "issue:45823",
+            "accept": false,
+            "reason": "Both are Gemma4 issues, but one is an FSDP2 shared_kv_states bug and the other is a device_map='auto' forward mismatch."
+          },
+          {
+            "left": "issue:43366",
+            "right": "issue:43531",
+            "accept": false,
+            "reason": "GGUF gpt-oss support and Qwen3-MoE sliding_window behavior are unrelated."
+          },
+          {
+            "left": "issue:43950",
+            "right": "issue:45834",
+            "accept": false,
+            "reason": "Silent non-persistent buffer corruption is unrelated to Kosmos2.5 OCR index errors."
+          },
+          {
+            "left": "issue:43550",
+            "right": "issue:44910",
+            "accept": false,
+            "reason": "Bamba torch.compile/SDPA failure and Qwen3.5 flash-attention memory access are different bugs."
+          },
+          {
+            "left": "issue:43645",
+            "right": "issue:43646",
+            "accept": true,
+            "reason": "Same custom model initialization regression in Transformers 5.0.0; 43645 just adds the Jupyter notebook context."
+          },
+          {
+            "left": "issue:43366",
+            "right": "issue:44589",
+            "accept": false,
+            "reason": "gpt-oss GGUF support and missing Float8 storage type are unrelated."
+          },
+          {
+            "left": "issue:43950",
+            "right": "issue:45325",
+            "accept": false,
+            "reason": "Non-persistent buffer corruption and Qwen2.5-VL rope scaling are different regressions."
+          },
+          {
+            "left": "issue:44560",
+            "right": "issue:44918",
+            "accept": false,
+            "reason": "Both mention Qwen3.5, but one is a video-input regression and the other is unpacking embeddings with TRL SFT trainer."
+          },
+          {
+            "left": "issue:43824",
+            "right": "issue:43957",
+            "accept": false,
+            "reason": "Missing Qwen2_5_VL export and meta-device loading failures are different issues."
+          },
+          {
+            "left": "issue:42617",
+            "right": "issue:43366",
+            "accept": false,
+            "reason": "A 3d_parallel.py runtime problem is unrelated to GGUF gpt-oss support."
+          },
+          {
+            "left": "issue:42915",
+            "right": "issue:43957",
+            "accept": false,
+            "reason": "Qwen3Moe FP8 loading and meta-device model loading are unrelated."
+          },
+          {
+            "left": "issue:36010",
+            "right": "issue:42915",
+            "accept": false,
+            "reason": "GenerationMixin import failure and Qwen3Moe FP8 loading are unrelated."
+          },
+          {
+            "left": "issue:44661",
+            "right": "issue:45823",
+            "accept": false,
+            "reason": "Tokenizer-mapping failure when adding a new model is unrelated to Gemma4 PLE device mismatch."
+          },
+          {
+            "left": "issue:38175",
+            "right": "issue:43994",
+            "accept": false,
+            "reason": "Both concern siglip2, but one is zero-probabilities behavior and the other is nonsensical outputs with AutoModel/pipeline; not clearly the same bug."
+          },
+          {
+            "left": "issue:43632",
+            "right": "issue:43872",
+            "accept": true,
+            "reason": "Both stem from the Transformers v5 _is_hf_initialized change; 43872 is the bitsandbytes failure mode of that regression."
+          }
+        ]
+      },
+      "evaluator_result": {
+        "accept": true,
+        "feedback": "The analyst stayed conservative and the accepted duplicate pairs are well grounded by the packet titles: 43645/43646, 43872/44291, 41093/44805, and the Qwen video/DeepSpeed pairs are plausible within the same bug family. The summary also correctly avoids claiming the whole bucket is one coherent duplicate set."
+      },
+      "evaluator_used": true,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": true,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 5,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:36010",
+        "issue:38175",
+        "issue:39404",
+        "issue:39692",
+        "issue:41720",
+        "issue:42491",
+        "issue:42915",
+        "issue:43257",
+        "issue:43317",
+        "issue:43596",
+        "issue:43638",
+        "issue:43644",
+        "issue:43645",
+        "issue:43708",
+        "issue:43720",
+        "issue:43854",
+        "issue:43856",
+        "issue:43873",
+        "issue:43931",
+        "issue:43994",
+        "issue:44075",
+        "issue:44155",
+        "issue:44387",
+        "issue:44451",
+        "issue:44488",
+        "issue:44534",
+        "issue:44560",
+        "issue:44661",
+        "issue:44756",
+        "issue:44805",
+        "issue:44877",
+        "issue:44910",
+        "issue:44960",
+        "issue:44977",
+        "issue:44991",
+        "issue:45030",
+        "issue:45276",
+        "issue:45292",
+        "issue:45313",
+        "issue:45335",
+        "issue:45357",
+        "issue:45362",
+        "issue:45406",
+        "issue:45488",
+        "issue:45584",
+        "issue:45588",
+        "issue:45812"
+      ],
+      "soft_pairs": [
+        "issue:44155|issue:45584",
+        "issue:44877|issue:45030",
+        "issue:44560|issue:44805",
+        "issue:43638|issue:44661",
+        "issue:43317|issue:44756",
+        "issue:43596|issue:44661",
+        "issue:44661|issue:44805",
+        "issue:38175|issue:39692",
+        "issue:43644|issue:44534",
+        "issue:45313|issue:45357",
+        "issue:45488|issue:45812",
+        "issue:44960|issue:44977",
+        "issue:44451|issue:44488",
+        "issue:42491|issue:43720",
+        "issue:43854|issue:43931",
+        "issue:41720|issue:42915",
+        "issue:43708|issue:44075",
+        "issue:43257|issue:43854",
+        "issue:42491|issue:45276",
+        "issue:43856|issue:43873",
+        "issue:42915|issue:44756",
+        "issue:42491|issue:43931",
+        "issue:39692|issue:43994",
+        "issue:43856|issue:44387",
+        "issue:43644|issue:43645",
+        "issue:44910|issue:45588",
+        "issue:45362|issue:45406",
+        "issue:42491|issue:45292",
+        "issue:44488|issue:44991",
+        "issue:45276|issue:45335",
+        "issue:36010|issue:39404"
+      ],
+      "prepared_review_unit_hash": "063b2ca454b689f48008dd2c5911e3b451ec8d83734f6fcebac87873c2f5fe0c",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 47,
+        "item_count": 47,
+        "soft_pair_count": 31,
+        "serialized_chars": 36407,
+        "estimated_input_tokens": 9102,
+        "estimated_eval_tokens": 18460
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "This cluster is mostly heterogeneous, with only a couple of clear duplicate pairs. The strongest duplicate is the Transformers v5 non-persistent-buffer regression; a second plausible duplicate is the resize_token_embeddings failure across model variants.",
+        "confidence": 0.79,
+        "canonical_issue_reason": "No cluster-wide canonical issue: the node set spans many unrelated bugs. Among the duplicate candidates, the non-persistent-buffer regression is the clearest shared issue.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "44534 is the best representative issue: it states the Transformers v5 regression succinctly and broadly, and is the clearest canonical wording of the duplicate pair with 43644.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:44155",
+            "right": "issue:45584",
+            "accept": false,
+            "reason": "Different audio bugs: batched inference token leakage vs Whisper empty-transcription handling."
+          },
+          {
+            "left": "issue:44877",
+            "right": "issue:45030",
+            "accept": false,
+            "reason": "Both are config-validation failures, but for different models and different failure modes."
+          },
+          {
+            "left": "issue:44560",
+            "right": "issue:44805",
+            "accept": false,
+            "reason": "Unrelated symptoms: Qwen3-VL video StopIteration vs a generic mask/tensor shape mismatch."
+          },
+          {
+            "left": "issue:43638",
+            "right": "issue:44661",
+            "accept": false,
+            "reason": "Different code paths: ZeRO-3/Bert initialization vs tokenizer mapping in add-new-model-like."
+          },
+          {
+            "left": "issue:43317",
+            "right": "issue:44756",
+            "accept": false,
+            "reason": "Both touch loading/offload memory, but one is a dequantized device_map failure and the other is a Strix Halo mmap OOM."
+          },
+          {
+            "left": "issue:43596",
+            "right": "issue:44661",
+            "accept": false,
+            "reason": "Different bugs: ZeRO-3 Bert index error vs tokenizer mapping-name handling."
+          },
+          {
+            "left": "issue:44661",
+            "right": "issue:44805",
+            "accept": false,
+            "reason": "No shared underlying bug; tokenizer mapping logic and mask-shape indexing are unrelated."
+          },
+          {
+            "left": "issue:38175",
+            "right": "issue:39692",
+            "accept": false,
+            "reason": "Same model family, but one is zero probabilities and the other is a doc-example/model-processor mismatch plus quantization failure."
+          },
+          {
+            "left": "issue:43644",
+            "right": "issue:44534",
+            "accept": true,
+            "reason": "Same Transformers v5 regression: non-persistent buffers are filled with junk."
+          },
+          {
+            "left": "issue:45313",
+            "right": "issue:45357",
+            "accept": false,
+            "reason": "Both are Qwen3.5 regressions, but one is ZeRO-3 weight loading and the other is bad save_pretrained visual-encoder keys."
+          },
+          {
+            "left": "issue:45488",
+            "right": "issue:45812",
+            "accept": false,
+            "reason": "Both are tokenizer regressions, but they affect different model families and different concrete tokenization paths."
+          },
+          {
+            "left": "issue:44960",
+            "right": "issue:44977",
+            "accept": false,
+            "reason": "Different models and failures; GLM5 is not the same bug as Qwen3.5 flash-attention generation."
+          },
+          {
+            "left": "issue:44451",
+            "right": "issue:44488",
+            "accept": false,
+            "reason": "Both are model-loading failures, but there is not enough evidence they share the same root cause."
+          },
+          {
+            "left": "issue:42491",
+            "right": "issue:43720",
+            "accept": false,
+            "reason": "Different domains: Qwen3 MoE LoRA compatibility vs BitNet packed-weight loading."
+          },
+          {
+            "left": "issue:43854",
+            "right": "issue:43931",
+            "accept": false,
+            "reason": "Different model-loading bugs for different architectures and failure types."
+          },
+          {
+            "left": "issue:41720",
+            "right": "issue:42915",
+            "accept": false,
+            "reason": "Both involve Qwen3-family failures, but one is device_map/cuda assert and the other is FineGrainedFP8Config."
+          },
+          {
+            "left": "issue:43708",
+            "right": "issue:44075",
+            "accept": false,
+            "reason": "Unrelated training scheduling vs optimizer-argument usage bug."
+          },
+          {
+            "left": "issue:43257",
+            "right": "issue:43854",
+            "accept": false,
+            "reason": "Different subsystems and models: Qwen3 MoE accelerate/deepspeed loading vs GLM-4.7 test loading."
+          },
+          {
+            "left": "issue:42491",
+            "right": "issue:45276",
+            "accept": false,
+            "reason": "Different issues: Qwen3 MoE LoRA compatibility vs Gemma4 embedding resizing."
+          },
+          {
+            "left": "issue:43856",
+            "right": "issue:43873",
+            "accept": false,
+            "reason": "Both mention memory, but one is Qwen3 MoE training efficiency and the other is quantization offload behavior."
+          },
+          {
+            "left": "issue:42915",
+            "right": "issue:44756",
+            "accept": false,
+            "reason": "Different memory-related failures with no shared concrete code path."
+          },
+          {
+            "left": "issue:42491",
+            "right": "issue:43931",
+            "accept": false,
+            "reason": "Different model families and bugs: Qwen3 MoE LoRA compatibility vs Qwen3-VL shape mismatch."
+          },
+          {
+            "left": "issue:39692",
+            "right": "issue:43994",
+            "accept": false,
+            "reason": "Same SigLIP2 family, but the reported bug details differ enough that this does not look like one duplicate."
+          },
+          {
+            "left": "issue:43856",
+            "right": "issue:44387",
+            "accept": false,
+            "reason": "Both are memory problems, but one is training inefficiency and the other is int4 reserved-memory OOM regression."
+          },
+          {
+            "left": "issue:43644",
+            "right": "issue:43645",
+            "accept": false,
+            "reason": "Both are v5 regressions, but one is buffer initialization and the other is notebook custom-model initialization."
+          },
+          {
+            "left": "issue:44910",
+            "right": "issue:45588",
+            "accept": false,
+            "reason": "Different flash-attention bugs: packed-sequence position_ids vs sink-less-model s_aux handling."
+          },
+          {
+            "left": "issue:45362",
+            "right": "issue:45406",
+            "accept": false,
+            "reason": "Different models and failures: Qwen3.5 chat crash vs Gemma4Processor attribute error in serve."
+          },
+          {
+            "left": "issue:42491",
+            "right": "issue:45292",
+            "accept": false,
+            "reason": "Different model-specific loading/compatibility bug vs generic resize_token_embeddings behavior."
+          },
+          {
+            "left": "issue:44488",
+            "right": "issue:44991",
+            "accept": false,
+            "reason": "Both are tokenizer-loading regressions, but the evidence only shows separate model-specific failures."
+          },
+          {
+            "left": "issue:45276",
+            "right": "issue:45335",
+            "accept": true,
+            "reason": "Same underlying resize_token_embeddings regression: embeddings are not updated on model-specific output/decoder embeddings."
+          },
+          {
+            "left": "issue:36010",
+            "right": "issue:39404",
+            "accept": false,
+            "reason": "Unrelated issues: a GenerationMixin import error vs Whisper pipeline return_language behavior."
+          }
+        ]
+      },
+      "evaluator_result": {
+        "accept": true,
+        "feedback": "Grounded and conservative overall. The cluster is clearly heterogeneous, and the two accepted duplicate pairs are well supported by the issue titles: 43644/44534 share the same Transformers v5 non-persistent-buffer regression, and 45276/45335 share the resize_token_embeddings regression across model variants. I did not see any soft-edge verdicts that overstate the evidence."
+      },
+      "evaluator_used": true,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": true,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 6,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:39692",
+        "issue:41720",
+        "issue:42947",
+        "issue:43257",
+        "issue:43335",
+        "issue:43381",
+        "issue:43404",
+        "issue:43425",
+        "issue:43454",
+        "issue:43645",
+        "issue:43646",
+        "issue:43708",
+        "issue:43828",
+        "issue:43854",
+        "issue:43873",
+        "issue:43906",
+        "issue:43957",
+        "issue:44155",
+        "issue:44164",
+        "issue:44291",
+        "issue:44292",
+        "issue:44387",
+        "issue:44402",
+        "issue:44451",
+        "issue:44479",
+        "issue:44488",
+        "issue:44530",
+        "issue:44589",
+        "issue:44683",
+        "issue:44756",
+        "issue:44863",
+        "issue:44877",
+        "issue:44912",
+        "issue:44928",
+        "issue:44960",
+        "issue:45216",
+        "issue:45276",
+        "issue:45292",
+        "issue:45335",
+        "issue:45588",
+        "issue:45663",
+        "issue:45684",
+        "issue:45698",
+        "issue:45704",
+        "issue:45759",
+        "issue:45823",
+        "issue:45859"
+      ],
+      "soft_pairs": [
+        "issue:43404|issue:43454",
+        "issue:44164|issue:44291",
+        "issue:44402|issue:44488",
+        "issue:43425|issue:43957",
+        "issue:39692|issue:44960",
+        "issue:43335|issue:45663",
+        "issue:42947|issue:45704",
+        "issue:43873|issue:44387",
+        "issue:45663|issue:45759",
+        "issue:45663|issue:45859",
+        "issue:44155|issue:45823",
+        "issue:43854|issue:45216",
+        "issue:44387|issue:44756",
+        "issue:44291|issue:44756",
+        "issue:44292|issue:44912",
+        "issue:42947|issue:43381",
+        "issue:45684|issue:45698",
+        "issue:45276|issue:45292",
+        "issue:43381|issue:44928",
+        "issue:44164|issue:44756",
+        "issue:41720|issue:44155",
+        "issue:44479|issue:44877",
+        "issue:43646|issue:43708",
+        "issue:45588|issue:45663",
+        "issue:43854|issue:44960",
+        "issue:43957|issue:44756",
+        "issue:45292|issue:45335",
+        "issue:44402|issue:44451",
+        "issue:43828|issue:43957",
+        "issue:43645|issue:44164",
+        "issue:43257|issue:44863",
+        "issue:43425|issue:43828",
+        "issue:44589|issue:44756",
+        "issue:44683|issue:45823",
+        "issue:43906|issue:44530"
+      ],
+      "prepared_review_unit_hash": "bdece9ceb2d966be418903dee166308c8f9e29bdd0365a29497661cb2bb95d82",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 47,
+        "item_count": 47,
+        "soft_pair_count": 35,
+        "serialized_chars": 37684,
+        "estimated_input_tokens": 9421,
+        "estimated_eval_tokens": 19098
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "The set is mostly a grab bag of unrelated model-loading, quantization, and training bugs. The only clear duplicate-style matches are the `resize_token_embeddings` regressions, where the generic issue and the model-specific reports look like the same underlying bug.",
+        "confidence": 0.87,
+        "canonical_issue_reason": "issue:45292 is the best canonical issue because it states the underlying `resize_token_embeddings` bug in the most general form (`output_embeddings` not updated), while the other matching reports are model-specific variants of the same failure.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "issue:45292 is the best global issue representative for the duplicate cluster, since it captures the shared `resize_token_embeddings` regression more broadly than the model-specific variants.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:43404",
+            "right": "issue:43454",
+            "accept": false,
+            "reason": "Both are lm_head tying bugs, but they hit different model classes and likely different code paths; too broad to treat as the same duplicate."
+          },
+          {
+            "left": "issue:44164",
+            "right": "issue:44291",
+            "accept": false,
+            "reason": "One is about `extra_state` in save/load, the other about `init_empty_weights` passing `_is_hf_initialized`; related area, different bug."
+          },
+          {
+            "left": "issue:44402",
+            "right": "issue:44488",
+            "accept": false,
+            "reason": "Both are model loading failures, but the titles point to different models and different root causes."
+          },
+          {
+            "left": "issue:43425",
+            "right": "issue:43957",
+            "accept": false,
+            "reason": "Torch version incompatibility is not the same as meta-device loading failures; different failure modes."
+          },
+          {
+            "left": "issue:39692",
+            "right": "issue:44960",
+            "accept": false,
+            "reason": "Completely different models and symptoms; no shared code-path evidence."
+          },
+          {
+            "left": "issue:43335",
+            "right": "issue:45663",
+            "accept": false,
+            "reason": "Different architectures and failures; only a superficial theme of configuration/runtime bugs."
+          },
+          {
+            "left": "issue:42947",
+            "right": "issue:45704",
+            "accept": false,
+            "reason": "Gradient checkpointing/LoRA is unrelated to a T5 RMSNorm memory leak."
+          },
+          {
+            "left": "issue:43873",
+            "right": "issue:44387",
+            "accept": false,
+            "reason": "Both involve quantization and OOMs, but one is offloading behavior and the other is reserved-memory growth under int4; not the same bug."
+          },
+          {
+            "left": "issue:45663",
+            "right": "issue:45759",
+            "accept": false,
+            "reason": "Different features: FSDP2 shared_kv_states rebuild versus multimodal config unwrapping."
+          },
+          {
+            "left": "issue:45663",
+            "right": "issue:45859",
+            "accept": false,
+            "reason": "Different Qwen/Gemma runtime paths and different failures; not the same underlying bug."
+          },
+          {
+            "left": "issue:44155",
+            "right": "issue:45823",
+            "accept": false,
+            "reason": "Both are forward-time bugs in multimodal models, but the reported mechanisms are different and not mergeable as one fix."
+          },
+          {
+            "left": "issue:43854",
+            "right": "issue:45216",
+            "accept": false,
+            "reason": "Different model families and different bugs: unit-test loading vs incorrect save_pretrained checkpoint handling."
+          },
+          {
+            "left": "issue:44387",
+            "right": "issue:44756",
+            "accept": false,
+            "reason": "Both mention OOM, but the causes are unrelated."
+          },
+          {
+            "left": "issue:44291",
+            "right": "issue:44756",
+            "accept": false,
+            "reason": "`init_empty_weights` loading error and Strix Halo mmap OOM are unrelated."
+          },
+          {
+            "left": "issue:44292",
+            "right": "issue:44912",
+            "accept": false,
+            "reason": "Different quantization formats, models, and symptoms; not the same concrete issue."
+          },
+          {
+            "left": "issue:42947",
+            "right": "issue:43381",
+            "accept": false,
+            "reason": "One says checkpointing is ineffective, the other says it cannot be used in eval mode; different behaviors."
+          },
+          {
+            "left": "issue:45684",
+            "right": "issue:45698",
+            "accept": false,
+            "reason": "Both concern custom-model save/load, but one is permissions propagation and the other is wrong-module reloading."
+          },
+          {
+            "left": "issue:45276",
+            "right": "issue:45292",
+            "accept": true,
+            "reason": "Same underlying `resize_token_embeddings` bug; 45276 is the Gemma4-specific manifestation of the broader 45292 report."
+          },
+          {
+            "left": "issue:43381",
+            "right": "issue:44928",
+            "accept": false,
+            "reason": "Gradient checkpointing in eval mode is unrelated to the Qwen3.5 RLHF NaN/SDPA fallback issue."
+          },
+          {
+            "left": "issue:44164",
+            "right": "issue:44756",
+            "accept": false,
+            "reason": "Save/load `extra_state` handling is unrelated to mmap-induced OOM."
+          },
+          {
+            "left": "issue:41720",
+            "right": "issue:44155",
+            "accept": false,
+            "reason": "Device-map CUDA assert and AudioFlamingo3 token leak are different bugs."
+          },
+          {
+            "left": "issue:44479",
+            "right": "issue:44877",
+            "accept": false,
+            "reason": "Both are regression-style multimodal issues, but they affect different models and different subsystems."
+          },
+          {
+            "left": "issue:43646",
+            "right": "issue:43708",
+            "accept": false,
+            "reason": "Custom model initialization breakage and Trainer checkpoint step calculation are unrelated."
+          },
+          {
+            "left": "issue:45588",
+            "right": "issue:45663",
+            "accept": false,
+            "reason": "Flash-attention sink handling and Gemma4 FSDP2 shared state rebuilding are unrelated."
+          },
+          {
+            "left": "issue:43854",
+            "right": "issue:44960",
+            "accept": false,
+            "reason": "Both are GLM-related, but one is a specific load/test issue and the other is a broad GLM5 report."
+          },
+          {
+            "left": "issue:45292",
+            "right": "issue:45335",
+            "accept": true,
+            "reason": "Same `resize_token_embeddings` regression; 45335 is the T5Gemma-specific manifestation of the broader output-embedding bug."
+          },
+          {
+            "left": "issue:44402",
+            "right": "issue:44451",
+            "accept": false,
+            "reason": "Different model load failures with no evidence of a shared root cause."
+          },
+          {
+            "left": "issue:43828",
+            "right": "issue:43957",
+            "accept": false,
+            "reason": "Autocast dtype mismatch and meta-device loading failures are different issues."
+          },
+          {
+            "left": "issue:43645",
+            "right": "issue:44164",
+            "accept": false,
+            "reason": "Jupyter/custom-model initialization breakage is not the same as `extra_state` save/load handling."
+          },
+          {
+            "left": "issue:43257",
+            "right": "issue:44863",
+            "accept": false,
+            "reason": "Both are checkpoint/loading problems, but the models and failure mechanisms differ."
+          },
+          {
+            "left": "issue:43425",
+            "right": "issue:43828",
+            "accept": false,
+            "reason": "Torch version incompatibility is not the same as the Phi-tiny-MoE autocast dtype mismatch."
+          },
+          {
+            "left": "issue:44589",
+            "right": "issue:44756",
+            "accept": false,
+            "reason": "Float8 storage deserialization and mmap OOM are unrelated."
+          },
+          {
+            "left": "issue:44683",
+            "right": "issue:45823",
+            "accept": false,
+            "reason": "Compiled flex_attention on newer torch and Gemma4 device mismatch are unrelated."
+          },
+          {
+            "left": "issue:43906",
+            "right": "issue:44530",
+            "accept": false,
+            "reason": "`43906` is only a reproduction of an earlier issue and does not clearly match the specific Qwen3.5 `linear_attention` crash."
+          }
+        ]
+      },
+      "evaluator_result": {
+        "accept": true,
+        "feedback": "The summary is grounded in the packet: issue:45292 is a reasonable canonical representative for the resize_token_embeddings regression, and the accepted soft edges to issue:45276 and issue:45335 are conservative given the very similar titles and the shared output_embedding/decoder.embed_tokens symptom. The other pairwise rejections are generally cautious and do not overstate evidence."
+      },
+      "evaluator_used": true,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": true,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 7,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:32090",
+        "issue:36032",
+        "issue:39404",
+        "issue:40990",
+        "issue:42915",
+        "issue:42947",
+        "issue:43299",
+        "issue:43421",
+        "issue:43425",
+        "issue:43452",
+        "issue:43582",
+        "issue:43596",
+        "issue:43611",
+        "issue:43632",
+        "issue:43644",
+        "issue:43645",
+        "issue:43716",
+        "issue:43856",
+        "issue:43901",
+        "issue:43931",
+        "issue:43994",
+        "issue:44079",
+        "issue:44164",
+        "issue:44206",
+        "issue:44220",
+        "issue:44265",
+        "issue:44479",
+        "issue:44488",
+        "issue:44509",
+        "issue:44530",
+        "issue:44589",
+        "issue:44661",
+        "issue:44671",
+        "issue:44683",
+        "issue:44737",
+        "issue:44756",
+        "issue:44805",
+        "issue:44863",
+        "issue:44898",
+        "issue:44912",
+        "issue:44918",
+        "issue:44991",
+        "issue:45081",
+        "issue:45084",
+        "issue:45701",
+        "issue:45759",
+        "issue:45823",
+        "issue:45859"
+      ],
+      "soft_pairs": [
+        "issue:39404|issue:45701",
+        "issue:43425|issue:45823",
+        "issue:43611|issue:43645",
+        "issue:43425|issue:43716",
+        "issue:42915|issue:44863",
+        "issue:44530|issue:45084",
+        "issue:43901|issue:44509",
+        "issue:39404|issue:44912",
+        "issue:44265|issue:44898",
+        "issue:43299|issue:43931",
+        "issue:44479|issue:44805",
+        "issue:43425|issue:44918",
+        "issue:44589|issue:45084",
+        "issue:43856|issue:44756",
+        "issue:40990|issue:42947",
+        "issue:39404|issue:44671",
+        "issue:43994|issue:44079",
+        "issue:44164|issue:45759",
+        "issue:36032|issue:43452",
+        "issue:43596|issue:44683",
+        "issue:44671|issue:44991",
+        "issue:44991|issue:45081",
+        "issue:44737|issue:45859",
+        "issue:32090|issue:39404",
+        "issue:44164|issue:44479",
+        "issue:44488|issue:45081",
+        "issue:44661|issue:44683",
+        "issue:40990|issue:43421",
+        "issue:43582|issue:43632",
+        "issue:43611|issue:43644",
+        "issue:44206|issue:44220",
+        "issue:43716|issue:45084"
+      ],
+      "prepared_review_unit_hash": "ed6af37e5c8fe762aa29e551cc635bf57d571b69fd9db788c6fe29b86914d769",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 32,
+        "serialized_chars": 37279,
+        "estimated_input_tokens": 9320,
+        "estimated_eval_tokens": 18896
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "The cluster is heterogeneous: most items are unrelated Transformers regressions or feature requests, and none of the soft pairs look like the same underlying bug/change.",
+        "confidence": 0.63,
+        "canonical_issue_reason": "issue:43611 is the best representative anchor because it is a concrete v5 model-loading regression with a clear code path and broader scope than the more model-specific reports, though the overall cluster is not a true duplicate set.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "issue:43611 is the strongest issue to keep as the cluster anchor: it is specific enough to be actionable, but broad enough to represent the dominant v5 loading-regression theme in this noisy group.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:39404",
+            "right": "issue:45701",
+            "accept": false,
+            "reason": "Whisper pipeline return_language regression and a tokenization-behavior report are different problems and code paths."
+          },
+          {
+            "left": "issue:43425",
+            "right": "issue:45823",
+            "accept": false,
+            "reason": "Torch 2.10 incompatibility and Gemma4 device-mismatch during forward are unrelated failures."
+          },
+          {
+            "left": "issue:43611",
+            "right": "issue:43645",
+            "accept": false,
+            "reason": "base_model_prefix loading regression and Jupyter custom-model initialization issues are distinct bugs."
+          },
+          {
+            "left": "issue:43425",
+            "right": "issue:43716",
+            "accept": false,
+            "reason": "A torch version compatibility issue is not the same as a multimodal dtype mismatch."
+          },
+          {
+            "left": "issue:42915",
+            "right": "issue:44863",
+            "accept": false,
+            "reason": "Different model families and different loading failures; no shared concrete bug."
+          },
+          {
+            "left": "issue:44530",
+            "right": "issue:45084",
+            "accept": false,
+            "reason": "PagedAttentionCache linear_attention crash and template compilation error are separate code paths."
+          },
+          {
+            "left": "issue:43901",
+            "right": "issue:44509",
+            "accept": false,
+            "reason": "Docs mentioning return_all_scores and docs mentioning removed pipeline tasks are different documentation issues."
+          },
+          {
+            "left": "issue:39404",
+            "right": "issue:44912",
+            "accept": false,
+            "reason": "Whisper return_language behavior and MXFP4 quantization fallback are unrelated."
+          },
+          {
+            "left": "issue:44265",
+            "right": "issue:44898",
+            "accept": false,
+            "reason": "torch.export / torch_compilable_check failure and Perceiver non-default resolution failure are different bugs."
+          },
+          {
+            "left": "issue:43299",
+            "right": "issue:43931",
+            "accept": false,
+            "reason": "Qwen3VL-MoE loading breakage and Qwen3-VL-30B weight-shape mismatch are not the same issue."
+          },
+          {
+            "left": "issue:44479",
+            "right": "issue:44805",
+            "accept": false,
+            "reason": "Video-input regression for Qwen VL models and a mask-shape IndexError are not the same underlying bug."
+          },
+          {
+            "left": "issue:43425",
+            "right": "issue:44918",
+            "accept": false,
+            "reason": "Torch compatibility and Qwen3.5 embedding unpacking in TRL are unrelated."
+          },
+          {
+            "left": "issue:44589",
+            "right": "issue:45084",
+            "accept": false,
+            "reason": "Float8 storage lookup failure and non-template-node compile error are different failures."
+          },
+          {
+            "left": "issue:43856",
+            "right": "issue:44756",
+            "accept": false,
+            "reason": "Qwen3 MoE training memory usage and mmap OOM on Strix Halo are different performance problems."
+          },
+          {
+            "left": "issue:40990",
+            "right": "issue:42947",
+            "accept": false,
+            "reason": "High perplexity on gpt-oss and gradient checkpointing ineffectiveness with LoRA are unrelated."
+          },
+          {
+            "left": "issue:39404",
+            "right": "issue:44671",
+            "accept": false,
+            "reason": "Whisper pipeline regression and CamemBERT masked-LM prediction regression are different model issues."
+          },
+          {
+            "left": "issue:43994",
+            "right": "issue:44079",
+            "accept": false,
+            "reason": "SigLIP2 bad outputs and ModelOutput key assignment with None are unrelated."
+          },
+          {
+            "left": "issue:44164",
+            "right": "issue:45759",
+            "accept": false,
+            "reason": "extra_state save/load handling and Qwen multimodal text_config unwrapping are different serialization paths."
+          },
+          {
+            "left": "issue:36032",
+            "right": "issue:43452",
+            "accept": false,
+            "reason": "Tokenizer add_special_tokens conflict and gguf_file loading breakage are distinct tokenizer/loading bugs."
+          },
+          {
+            "left": "issue:43596",
+            "right": "issue:44683",
+            "accept": false,
+            "reason": "DeepSpeed zero3 BertModel IndexError and compiled flex_attention torch-version failure are unrelated."
+          },
+          {
+            "left": "issue:44671",
+            "right": "issue:44991",
+            "accept": false,
+            "reason": "CamemBERT prediction regression and tokenizer loading failure for a different checkpoint are not the same bug."
+          },
+          {
+            "left": "issue:44991",
+            "right": "issue:45081",
+            "accept": false,
+            "reason": "Different tokenizer regressions with different failure modes and different affected models."
+          },
+          {
+            "left": "issue:44737",
+            "right": "issue:45859",
+            "accept": false,
+            "reason": "XLNet CPU device placement bug and CPU-offload incompatibility in Qwen3_5Moe are separate issues."
+          },
+          {
+            "left": "issue:32090",
+            "right": "issue:39404",
+            "accept": false,
+            "reason": "Trainer _gpu_broadcast_one NoneType error and Whisper return_language regression are unrelated."
+          },
+          {
+            "left": "issue:44164",
+            "right": "issue:44479",
+            "accept": false,
+            "reason": "Serialization extra_state handling and video input regression are different code paths."
+          },
+          {
+            "left": "issue:44488",
+            "right": "issue:45081",
+            "accept": false,
+            "reason": "Loading cjvt/sleng-bert and Mistral regex patching failures are distinct tokenizer issues."
+          },
+          {
+            "left": "issue:44661",
+            "right": "issue:44683",
+            "accept": false,
+            "reason": "Tokenizer mapping-name failure and compiled flex_attention failure are unrelated."
+          },
+          {
+            "left": "issue:40990",
+            "right": "issue:43421",
+            "accept": false,
+            "reason": "High perplexity on a model and runtime post-processor updates in TokenizersBackend are not the same issue."
+          },
+          {
+            "left": "issue:43582",
+            "right": "issue:43632",
+            "accept": false,
+            "reason": "Apple Silicon caching_allocator_warmup TypeError and _is_hf_initialized flag breakage are different failures."
+          },
+          {
+            "left": "issue:43611",
+            "right": "issue:43644",
+            "accept": false,
+            "reason": "Model loading with base_model_prefix and junk-filled non-persistent buffers are different regressions."
+          },
+          {
+            "left": "issue:44206",
+            "right": "issue:44220",
+            "accept": false,
+            "reason": "Unsupported center argument in LasrFeatureExtractor and _torch_extract_fbank_features are related to audio preprocessing but not clearly the same concrete bug."
+          },
+          {
+            "left": "issue:43716",
+            "right": "issue:45084",
+            "accept": false,
+            "reason": "Mistral-3 dtype mismatch and non-template-node compilation error are unrelated."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 8,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:34567",
+        "issue:36032",
+        "issue:40444",
+        "issue:42915",
+        "issue:43262",
+        "issue:43296",
+        "issue:43317",
+        "issue:43322",
+        "issue:43377",
+        "issue:43404",
+        "issue:43452",
+        "issue:43482",
+        "issue:43531",
+        "issue:43540",
+        "issue:43541",
+        "issue:43632",
+        "issue:43643",
+        "issue:43644",
+        "issue:43645",
+        "issue:43646",
+        "issue:43819",
+        "issue:43883",
+        "issue:43994",
+        "issue:44062",
+        "issue:44220",
+        "issue:44265",
+        "issue:44291",
+        "issue:44403",
+        "issue:44451",
+        "issue:44466",
+        "issue:44488",
+        "issue:44530",
+        "issue:44534",
+        "issue:44610",
+        "issue:44683",
+        "issue:44756",
+        "issue:44811",
+        "issue:44849",
+        "issue:44987",
+        "issue:44991",
+        "issue:45081",
+        "issue:45084",
+        "issue:45127",
+        "issue:45216",
+        "issue:45325",
+        "issue:45357",
+        "issue:45507",
+        "issue:45584"
+      ],
+      "soft_pairs": [
+        "issue:43883|issue:44488",
+        "issue:44488|issue:44987",
+        "issue:43643|issue:44530",
+        "issue:34567|issue:40444",
+        "issue:44811|issue:45584",
+        "issue:43644|issue:43646",
+        "issue:44451|issue:44987",
+        "issue:43541|issue:45084",
+        "issue:36032|issue:42915",
+        "issue:43452|issue:43482",
+        "issue:42915|issue:44530",
+        "issue:43317|issue:43482",
+        "issue:44849|issue:45325",
+        "issue:43883|issue:44991",
+        "issue:43404|issue:44466",
+        "issue:36032|issue:43645",
+        "issue:45084|issue:45507",
+        "issue:43262|issue:43377",
+        "issue:43646|issue:44534",
+        "issue:44987|issue:44991",
+        "issue:43994|issue:44610",
+        "issue:44403|issue:45081",
+        "issue:43322|issue:43540",
+        "issue:43377|issue:43819",
+        "issue:43645|issue:44534",
+        "issue:42915|issue:45084",
+        "issue:44451|issue:45081",
+        "issue:45216|issue:45357",
+        "issue:44265|issue:44610",
+        "issue:44987|issue:45081",
+        "issue:44220|issue:44683",
+        "issue:43404|issue:45127",
+        "issue:43296|issue:43531",
+        "issue:43632|issue:44756",
+        "issue:42915|issue:44291",
+        "issue:44062|issue:44987"
+      ],
+      "prepared_review_unit_hash": "c5a3b15716260e725c6382fc5e208214d22dd9e0dbb4f62050bd038363741a97",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 36,
+        "serialized_chars": 38560,
+        "estimated_input_tokens": 9640,
+        "estimated_eval_tokens": 19536
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "This cluster is mostly a collection of separate Transformers regressions around loading, tokenizer/model init, quantization, and model-specific edge cases. I did not find any soft pair that looks like the same concrete bug strongly enough to merge.",
+        "confidence": 0.84,
+        "canonical_issue_reason": "Issue 43646 is the broadest representative of the v5 model-initialization regression subgroup, so it is the best anchor if one must be chosen; however, the cluster is heterogeneous and not a true duplicate set.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "43646 is the most general issue title among the initialization-related reports and is the clearest representative of the closest mini-cluster, though still not a reliable duplicate canonical.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:43883",
+            "right": "issue:44488",
+            "accept": false,
+            "reason": "Both are model-loading failures, but for different models and different error paths; not the same bug."
+          },
+          {
+            "left": "issue:44488",
+            "right": "issue:44987",
+            "accept": false,
+            "reason": "Two separate checkpoint-loading failures for different models; no concrete shared code-path is evident."
+          },
+          {
+            "left": "issue:43643",
+            "right": "issue:44530",
+            "accept": false,
+            "reason": "Missing fields from AutoConfig with trust_remote_code is unrelated to PagedAttentionCache group-type handling."
+          },
+          {
+            "left": "issue:34567",
+            "right": "issue:40444",
+            "accept": false,
+            "reason": "Trainer token-count tracking and Qwen2.5-VL iterable dataset/image handling are unrelated."
+          },
+          {
+            "left": "issue:44811",
+            "right": "issue:45584",
+            "accept": false,
+            "reason": "Both involve Whisper token handling, but one is batch_decode skip_special_tokens behavior and the other is empty-transcription generation after align_special_tokens; different bugs."
+          },
+          {
+            "left": "issue:43644",
+            "right": "issue:43646",
+            "accept": false,
+            "reason": "Both are v5 initialization regressions, but one is non-persistent buffer serialization and the other is custom model initialization in notebooks; too different to merge."
+          },
+          {
+            "left": "issue:44451",
+            "right": "issue:44987",
+            "accept": false,
+            "reason": "Different model-loading failures for different checkpoints; no same underlying defect is shown."
+          },
+          {
+            "left": "issue:43541",
+            "right": "issue:45084",
+            "accept": false,
+            "reason": "Torch dynamo grouped_mm tracing failure and template-node compilation error are unrelated."
+          },
+          {
+            "left": "issue:36032",
+            "right": "issue:42915",
+            "accept": false,
+            "reason": "Tokenizer attribute conflict and Qwen3Moe FineGrainedFP8Config failure are different subsystems and failure modes."
+          },
+          {
+            "left": "issue:43452",
+            "right": "issue:43482",
+            "accept": false,
+            "reason": "Both mention GGUF loading, but one is API breakage around gguf_file and the other is a specific Qwen2.5-GGUF v5 failure; not clearly the same bug."
+          },
+          {
+            "left": "issue:42915",
+            "right": "issue:44530",
+            "accept": false,
+            "reason": "FineGrainedFP8Config loading failure and PagedAttentionCache linear_attention crash are unrelated."
+          },
+          {
+            "left": "issue:43317",
+            "right": "issue:43482",
+            "accept": false,
+            "reason": "device_map auto offload failure and GGUF model loading failure are different code paths."
+          },
+          {
+            "left": "issue:44849",
+            "right": "issue:45325",
+            "accept": false,
+            "reason": "Qwen3.5 hidden-states output bug and Qwen2.5-VL rope-index scaling bug are unrelated."
+          },
+          {
+            "left": "issue:43883",
+            "right": "issue:44991",
+            "accept": false,
+            "reason": "Both are tokenizer/model loading complaints, but for different models and different errors."
+          },
+          {
+            "left": "issue:43404",
+            "right": "issue:44466",
+            "accept": false,
+            "reason": "Both concern tied weights, but one is a model-specific untied lm_head bug and the other is serialization differing by device; not the same concrete issue."
+          },
+          {
+            "left": "issue:36032",
+            "right": "issue:43645",
+            "accept": false,
+            "reason": "Tokenizer method conflict and notebook custom-model initialization breakage are unrelated."
+          },
+          {
+            "left": "issue:45084",
+            "right": "issue:45507",
+            "accept": false,
+            "reason": "Template compilation error and GraniteMoEHybrid invalid method call are different failures."
+          },
+          {
+            "left": "issue:43262",
+            "right": "issue:43377",
+            "accept": false,
+            "reason": "Audio chat-template sampling-rate defaulting and MIMI batched-vs-single output mismatch are different bugs."
+          },
+          {
+            "left": "issue:43646",
+            "right": "issue:44534",
+            "accept": false,
+            "reason": "Both are v5 initialization-related, but one is about custom model init and the other about non-persistent buffers being filled with junk; related theme, not same bug."
+          },
+          {
+            "left": "issue:44987",
+            "right": "issue:44991",
+            "accept": false,
+            "reason": "Separate model-loading failures for different repositories and error causes."
+          },
+          {
+            "left": "issue:43994",
+            "right": "issue:44610",
+            "accept": false,
+            "reason": "Siglip2 nonsensical outputs and OmDet-Turbo image-size mismatch are different model issues."
+          },
+          {
+            "left": "issue:44403",
+            "right": "issue:45081",
+            "accept": false,
+            "reason": "Generic loading noise and Mistral regex patch crash are unrelated."
+          },
+          {
+            "left": "issue:43322",
+            "right": "issue:43540",
+            "accept": false,
+            "reason": "Llava Next segmentation fault and Qwen3OmniMoe video-processing ValueError are distinct multimodal loading/runtime bugs."
+          },
+          {
+            "left": "issue:43377",
+            "right": "issue:43819",
+            "accept": false,
+            "reason": "Padding-mask support in MIMI encoder and DAC from_latents STE mismatch are separate audio/model-consistency issues."
+          },
+          {
+            "left": "issue:43645",
+            "right": "issue:44534",
+            "accept": false,
+            "reason": "Custom-model initialization in notebooks is not the same as non-persistent buffer junk filling."
+          },
+          {
+            "left": "issue:42915",
+            "right": "issue:45084",
+            "accept": false,
+            "reason": "Qwen3Moe FP8 loading failure and template-node compilation error are unrelated."
+          },
+          {
+            "left": "issue:44451",
+            "right": "issue:45081",
+            "accept": false,
+            "reason": "Different model-loading failure and tokenizer regex patch crash; no shared concrete bug."
+          },
+          {
+            "left": "issue:45216",
+            "right": "issue:45357",
+            "accept": false,
+            "reason": "Both are Qwen3.5 save_pretrained regressions, but one is incorrect checkpoint saving broadly and the other is incorrect visual-encoder keys; related family but not the same specific bug."
+          },
+          {
+            "left": "issue:44265",
+            "right": "issue:44610",
+            "accept": false,
+            "reason": "torch.export exportability with torch_compilable_check is unrelated to OmDet-Turbo input resolution mismatch."
+          },
+          {
+            "left": "issue:44987",
+            "right": "issue:45081",
+            "accept": false,
+            "reason": "A model-loading regression and a tokenizer regex patch crash are different code paths."
+          },
+          {
+            "left": "issue:44220",
+            "right": "issue:44683",
+            "accept": false,
+            "reason": "Feature-extraction issue and compiled flex_attention failure are unrelated."
+          },
+          {
+            "left": "issue:43404",
+            "right": "issue:45127",
+            "accept": false,
+            "reason": "Both involve tied weights/LM head behavior, but one is missing tying in a specific model and the other is LoRA merge collapse with extended vocab; not the same bug."
+          },
+          {
+            "left": "issue:43296",
+            "right": "issue:43531",
+            "accept": false,
+            "reason": "PaddleOCR-VL load failure and Qwen3-MoE sliding_window bug are unrelated."
+          },
+          {
+            "left": "issue:43632",
+            "right": "issue:44756",
+            "accept": false,
+            "reason": "_is_hf_initialized flag regression and disabling mmap on Strix Halo to avoid OOM are unrelated."
+          },
+          {
+            "left": "issue:42915",
+            "right": "issue:44291",
+            "accept": false,
+            "reason": "FineGrainedFP8 loading and init_empty_weights unexpected _is_hf_initialized argument are different initialization/quantization issues."
+          },
+          {
+            "left": "issue:44062",
+            "right": "issue:44987",
+            "accept": false,
+            "reason": "AddedToken special-argument conflict and a model-loading failure for physical-intelligence/fast are unrelated."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 9,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:33666",
+        "issue:36032",
+        "issue:40444",
+        "issue:40990",
+        "issue:41553",
+        "issue:41720",
+        "issue:42175",
+        "issue:42915",
+        "issue:43296",
+        "issue:43317",
+        "issue:43377",
+        "issue:43421",
+        "issue:43425",
+        "issue:43482",
+        "issue:43525",
+        "issue:43531",
+        "issue:43582",
+        "issue:43611",
+        "issue:43632",
+        "issue:43646",
+        "issue:43742",
+        "issue:43792",
+        "issue:43818",
+        "issue:43856",
+        "issue:44079",
+        "issue:44291",
+        "issue:44403",
+        "issue:44462",
+        "issue:44479",
+        "issue:44534",
+        "issue:44568",
+        "issue:44589",
+        "issue:44683",
+        "issue:44811",
+        "issue:44829",
+        "issue:44877",
+        "issue:44991",
+        "issue:45020",
+        "issue:45081",
+        "issue:45325",
+        "issue:45357",
+        "issue:45479",
+        "issue:45491",
+        "issue:45538",
+        "issue:45542",
+        "issue:45663",
+        "issue:45812"
+      ],
+      "soft_pairs": [
+        "issue:36032|issue:43742",
+        "issue:42915|issue:44589",
+        "issue:43611|issue:44534",
+        "issue:43531|issue:44877",
+        "issue:41553|issue:43582",
+        "issue:43317|issue:43856",
+        "issue:44479|issue:45325",
+        "issue:43425|issue:44403",
+        "issue:40990|issue:44811",
+        "issue:43818|issue:45357",
+        "issue:42175|issue:45542",
+        "issue:44462|issue:45812",
+        "issue:36032|issue:41720",
+        "issue:43421|issue:44568",
+        "issue:44079|issue:45663",
+        "issue:43792|issue:44589",
+        "issue:43632|issue:44534",
+        "issue:44403|issue:44991",
+        "issue:44829|issue:45479",
+        "issue:43482|issue:45538",
+        "issue:44683|issue:45020",
+        "issue:33666|issue:40444",
+        "issue:43482|issue:43525",
+        "issue:44291|issue:44589",
+        "issue:40444|issue:45325",
+        "issue:43377|issue:45491",
+        "issue:41720|issue:45081",
+        "issue:42175|issue:43296",
+        "issue:43531|issue:44291",
+        "issue:43611|issue:43646"
+      ],
+      "prepared_review_unit_hash": "6aa89efb21ad358fa4f0c0caaf4b621b257a50a1e8621ffcd75b9ab44bc368ec",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 47,
+        "item_count": 47,
+        "soft_pair_count": 30,
+        "serialized_chars": 36181,
+        "estimated_input_tokens": 9046,
+        "estimated_eval_tokens": 18348
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "This cluster is mostly heterogeneous: several separate Transformers v5 regressions and model-specific loading/training bugs, but no soft pair looks like a true duplicate. The broadest umbrella issue is the open remote_code regression report.",
+        "confidence": 0.47,
+        "canonical_issue_reason": "issue:45020 is the broadest open issue here, covering recent-version breakage in model loading with remote_code, which is the closest umbrella for the loading/regression-heavy subset of this cluster.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "issue:45020 is the best single issue to keep as the cluster anchor because it is open, broad, and captures the common theme of recent Transformers regressions better than the model-specific reports.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:36032",
+            "right": "issue:43742",
+            "accept": false,
+            "reason": "Different models and failures: T5 tokenizer method conflict vs a key error while loading MobileLLM."
+          },
+          {
+            "left": "issue:42915",
+            "right": "issue:44589",
+            "accept": false,
+            "reason": "Both mention float8-related failures, but one is Qwen3Moe FP8 config handling and the other is a missing Float8 storage type; not the same bug."
+          },
+          {
+            "left": "issue:43611",
+            "right": "issue:44534",
+            "accept": false,
+            "reason": "Separate v5 regressions: base_model_prefix loading failure vs non-persistent buffer corruption."
+          },
+          {
+            "left": "issue:43531",
+            "right": "issue:44877",
+            "accept": false,
+            "reason": "Unrelated model/config bugs: Qwen3-MoE sliding_window issue vs granite_speech strict config loading."
+          },
+          {
+            "left": "issue:41553",
+            "right": "issue:43582",
+            "accept": false,
+            "reason": "Different subsystems entirely: tokenizer error message vs Apple Silicon allocator warmup TypeError."
+          },
+          {
+            "left": "issue:43317",
+            "right": "issue:43856",
+            "accept": false,
+            "reason": "One is device_map/offload loading failure; the other is Qwen3 MoE training memory inefficiency. Same broad area, different bug."
+          },
+          {
+            "left": "issue:44479",
+            "right": "issue:45325",
+            "accept": false,
+            "reason": "Both are Qwen VL-related, but one is a broad video-input regression and the other is a still-image temporal position_ids scaling bug."
+          },
+          {
+            "left": "issue:43425",
+            "right": "issue:44403",
+            "accept": false,
+            "reason": "Incompatible torch/backend issue vs an unrelated logging/noise issue during model loading."
+          },
+          {
+            "left": "issue:40990",
+            "right": "issue:44811",
+            "accept": false,
+            "reason": "Completely different problems: perplexity on GPT-OSS vs Whisper decode skipping special tokens."
+          },
+          {
+            "left": "issue:43818",
+            "right": "issue:45357",
+            "accept": false,
+            "reason": "Different model families and code paths: Video-LLaVA weight/temporal attention issue vs Qwen3.5 save_pretrained key regression."
+          },
+          {
+            "left": "issue:42175",
+            "right": "issue:45542",
+            "accept": false,
+            "reason": "Both involve TensorFlow packaging/backend, but one is missing TF from the torch extra and the other is an undefined tf backend when only tensorboard is installed."
+          },
+          {
+            "left": "issue:44462",
+            "right": "issue:45812",
+            "accept": false,
+            "reason": "Both are tokenizer regressions, but one ignores tokenizer.json and the other yields wrong Granite token IDs; not the same concrete failure."
+          },
+          {
+            "left": "issue:36032",
+            "right": "issue:41720",
+            "accept": false,
+            "reason": "Tokenizer method conflict vs Qwen3 auto device mapping cuda assert; unrelated bugs."
+          },
+          {
+            "left": "issue:43421",
+            "right": "issue:44568",
+            "accept": false,
+            "reason": "Both touch special-token handling, but one is runtime post-processor refresh and the other is add_special_tokens not adding BOS/EOS for a specific tokenizer."
+          },
+          {
+            "left": "issue:44079",
+            "right": "issue:45663",
+            "accept": false,
+            "reason": "ModelOutput key assignment bug vs Gemma4 FSDP2 shared_kv_states reconstruction error; different code paths."
+          },
+          {
+            "left": "issue:43792",
+            "right": "issue:44589",
+            "accept": false,
+            "reason": "Whisper runtime failure vs missing Float8 storage type; unrelated."
+          },
+          {
+            "left": "issue:43632",
+            "right": "issue:44534",
+            "accept": false,
+            "reason": "Different Transformers v5 regressions: _is_hf_initialized breakage vs junk-filled non-persistent buffers."
+          },
+          {
+            "left": "issue:44403",
+            "right": "issue:44991",
+            "accept": false,
+            "reason": "Broad loading noise report vs tokenizer loading failure for a specific model; not a duplicate."
+          },
+          {
+            "left": "issue:44829",
+            "right": "issue:45479",
+            "accept": false,
+            "reason": "Both are classification-training regressions, but one is flash_attention_3-induced degeneration and the other is a zero-loss problem from num_labels=1."
+          },
+          {
+            "left": "issue:43482",
+            "right": "issue:45538",
+            "accept": false,
+            "reason": "Qwen2.5-GGUF loading failure vs CLIPTokenizer model_max_length handling; unrelated."
+          },
+          {
+            "left": "issue:44683",
+            "right": "issue:45020",
+            "accept": false,
+            "reason": "Torch flex_attention compilation failure vs remote_code model loading regressions; different layers of the stack."
+          },
+          {
+            "left": "issue:33666",
+            "right": "issue:40444",
+            "accept": false,
+            "reason": "Both are Qwen-VL training issues, but multi-GPU training and iterable-dataset multi-image finetuning are distinct bugs."
+          },
+          {
+            "left": "issue:43482",
+            "right": "issue:43525",
+            "accept": false,
+            "reason": "Different loading failures: GGUF/v5 compatibility vs missing pad_token_id on Llama4Config."
+          },
+          {
+            "left": "issue:44291",
+            "right": "issue:44589",
+            "accept": false,
+            "reason": "Init_empty_weights/_is_hf_initialized argument issue vs Float8 storage lookup failure; not the same bug."
+          },
+          {
+            "left": "issue:40444",
+            "right": "issue:45325",
+            "accept": false,
+            "reason": "Different Qwen2.5-VL regressions: multi-image iterable dataset failure vs still-image temporal position scaling."
+          },
+          {
+            "left": "issue:43377",
+            "right": "issue:45491",
+            "accept": false,
+            "reason": "Different model families and symptoms: MIMI padding-mask issue vs Gemma3 sliding-window NaNs."
+          },
+          {
+            "left": "issue:41720",
+            "right": "issue:45081",
+            "accept": false,
+            "reason": "Qwen3 A800 cuda assert vs Mistral tokenizer regex patch crash; unrelated."
+          },
+          {
+            "left": "issue:42175",
+            "right": "issue:43296",
+            "accept": false,
+            "reason": "TensorFlow packaging/backend issue vs PaddleOCR-VL load failure in vLLM; separate causes."
+          },
+          {
+            "left": "issue:43531",
+            "right": "issue:44291",
+            "accept": false,
+            "reason": "Qwen3-MoE sliding_window bug vs init_empty_weights unexpected argument error; unrelated."
+          },
+          {
+            "left": "issue:43611",
+            "right": "issue:43646",
+            "accept": false,
+            "reason": "Both are v5 loading regressions, but one is base_model_prefix handling and the other is custom model initialization; too different to merge."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 10,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:35707",
+        "issue:39692",
+        "issue:40444",
+        "issue:40990",
+        "issue:41720",
+        "issue:42915",
+        "issue:43262",
+        "issue:43296",
+        "issue:43317",
+        "issue:43377",
+        "issue:43454",
+        "issue:43482",
+        "issue:43526",
+        "issue:43550",
+        "issue:43572",
+        "issue:43606",
+        "issue:43632",
+        "issue:43644",
+        "issue:43701",
+        "issue:43746",
+        "issue:43792",
+        "issue:43819",
+        "issue:43873",
+        "issue:43881",
+        "issue:43883",
+        "issue:43957",
+        "issue:43994",
+        "issue:44079",
+        "issue:44186",
+        "issue:44265",
+        "issue:44291",
+        "issue:44387",
+        "issue:44466",
+        "issue:44479",
+        "issue:44488",
+        "issue:44534",
+        "issue:44610",
+        "issue:44617",
+        "issue:44877",
+        "issue:44898",
+        "issue:45072",
+        "issue:45325",
+        "issue:45381",
+        "issue:45584",
+        "issue:45759",
+        "issue:45800",
+        "issue:45823",
+        "issue:45834"
+      ],
+      "soft_pairs": [
+        "issue:44265|issue:45072",
+        "issue:43632|issue:45072",
+        "issue:43454|issue:44466",
+        "issue:43994|issue:44898",
+        "issue:44079|issue:44534",
+        "issue:45325|issue:45834",
+        "issue:43792|issue:43957",
+        "issue:42915|issue:43482",
+        "issue:44079|issue:45072",
+        "issue:40444|issue:45381",
+        "issue:45325|issue:45381",
+        "issue:43644|issue:44079",
+        "issue:44387|issue:44617",
+        "issue:43881|issue:44488",
+        "issue:43296|issue:45800",
+        "issue:44479|issue:45823",
+        "issue:35707|issue:40444",
+        "issue:44291|issue:44877",
+        "issue:43317|issue:43873",
+        "issue:43606|issue:44265",
+        "issue:43701|issue:45325",
+        "issue:44186|issue:44265",
+        "issue:43606|issue:43819",
+        "issue:43262|issue:43746",
+        "issue:41720|issue:45823",
+        "issue:43526|issue:43883",
+        "issue:43454|issue:43550",
+        "issue:43377|issue:44610",
+        "issue:43377|issue:44898",
+        "issue:43454|issue:45072",
+        "issue:40990|issue:45584",
+        "issue:43572|issue:45759",
+        "issue:39692|issue:40444"
+      ],
+      "prepared_review_unit_hash": "e7f90e29b8ebe9c37644eb9fe77fc3070ae5229e6d700c1559880aedfb7195a2",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 33,
+        "serialized_chars": 37595,
+        "estimated_input_tokens": 9399,
+        "estimated_eval_tokens": 19054
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "Mostly a heterogeneous set of unrelated bug reports. I only found one plausible duplicate pair: the Qwen video-input regression reports (#44479 and #45381). All other soft edges look like different bugs or different code paths.",
+        "confidence": 0.71,
+        "canonical_issue_reason": "#44479 is the broadest and most representative report of the Qwen video-input regression; it covers multiple affected Qwen variants, while #45381 is the narrower qwen2.5-vl manifestation.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "#44479 is the best issue anchor here because it captures the most general version of the only likely duplicate bug in the set and is the clearest representative for triage.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:44265",
+            "right": "issue:45072",
+            "accept": false,
+            "reason": "Different bugs: export failures around torch_compilable_check vs bfloat16 dtype mismatches during inference."
+          },
+          {
+            "left": "issue:43632",
+            "right": "issue:45072",
+            "accept": false,
+            "reason": "Different issues: _is_hf_initialized / v5 initialization breakage vs dtype mismatch under bfloat16 inference."
+          },
+          {
+            "left": "issue:43454",
+            "right": "issue:44466",
+            "accept": false,
+            "reason": "Related theme (tied weights) but different code paths: one is failing to tie lm_head weights, the other is inconsistent serialization behavior by device."
+          },
+          {
+            "left": "issue:43994",
+            "right": "issue:44898",
+            "accept": false,
+            "reason": "Different models and failure modes: SigLIP2 AutoModel/pipeline nonsense outputs vs Perceiver non-default-resolution failure."
+          },
+          {
+            "left": "issue:44079",
+            "right": "issue:44534",
+            "accept": false,
+            "reason": "Different underlying bugs: ModelOutput key assignment vs non-persistent buffer junk in v5."
+          },
+          {
+            "left": "issue:45325",
+            "right": "issue:45834",
+            "accept": false,
+            "reason": "Different Qwen-family problems: still-image rope scaling bug vs Kosmos2.5 long-OCR index error."
+          },
+          {
+            "left": "issue:43792",
+            "right": "issue:43957",
+            "accept": false,
+            "reason": "Both involve model loading, but the concrete failures differ: whisper runtime failure vs meta-device loading breakage in v5."
+          },
+          {
+            "left": "issue:42915",
+            "right": "issue:43482",
+            "accept": false,
+            "reason": "Different problems: FineGrainedFP8Config failure vs GGUF loading failure in transformers v5."
+          },
+          {
+            "left": "issue:44079",
+            "right": "issue:45072",
+            "accept": false,
+            "reason": "Different bugs: ModelOutput key handling vs dtype mismatch during bfloat16 inference."
+          },
+          {
+            "left": "issue:40444",
+            "right": "issue:45381",
+            "accept": false,
+            "reason": "Both are Qwen2.5-VL related, but one is iterable-dataset multi-image finetuning and the other is video position-id computation."
+          },
+          {
+            "left": "issue:45325",
+            "right": "issue:45381",
+            "accept": false,
+            "reason": "Both concern Qwen2.5-VL positioning, but one is still-image temporal scaling and the other is video-input vision_position_ids."
+          },
+          {
+            "left": "issue:43644",
+            "right": "issue:44079",
+            "accept": false,
+            "reason": "Different regressions: non-persistent buffer junk vs ModelOutput key assignment."
+          },
+          {
+            "left": "issue:44387",
+            "right": "issue:44617",
+            "accept": false,
+            "reason": "Both mention OOM, but one is increased reserved memory under int4 quantization and the other is a model-specific video OOM."
+          },
+          {
+            "left": "issue:43881",
+            "right": "issue:44488",
+            "accept": false,
+            "reason": "Both are loading failures, but they affect unrelated models with no shared concrete code path evident."
+          },
+          {
+            "left": "issue:43296",
+            "right": "issue:45800",
+            "accept": false,
+            "reason": "Different layers and failures: PaddleOCR-VL load in vLLM vs torch/transformers version incompatibility."
+          },
+          {
+            "left": "issue:44479",
+            "right": "issue:45823",
+            "accept": false,
+            "reason": "Different bugs: Qwen video-input regression vs Gemma4 PLE device mismatch under device_map=auto."
+          },
+          {
+            "left": "issue:35707",
+            "right": "issue:40444",
+            "accept": false,
+            "reason": "Different generation/training scenarios and model families; no shared concrete bug path."
+          },
+          {
+            "left": "issue:44291",
+            "right": "issue:44877",
+            "accept": false,
+            "reason": "Different loading/config issues: unexpected _is_hf_initialized arg vs strict config blocking granite_speech config."
+          },
+          {
+            "left": "issue:43317",
+            "right": "issue:43873",
+            "accept": false,
+            "reason": "Related offload/quantization area, but one is dequantized model loading on gpu+cpu offload and the other is a broader offloading complaint."
+          },
+          {
+            "left": "issue:43606",
+            "right": "issue:44265",
+            "accept": false,
+            "reason": "Different failure modes and subsystems: CPU offload device mismatch vs torch.export / torch_compilable_check export failure."
+          },
+          {
+            "left": "issue:43701",
+            "right": "issue:45325",
+            "accept": false,
+            "reason": "Completely different problems: checkpoint resume key mismatch vs Qwen2.5-VL rope index scaling."
+          },
+          {
+            "left": "issue:44186",
+            "right": "issue:44265",
+            "accept": false,
+            "reason": "Tokenizer padding/truncation crash vs torch.export failure; unrelated code paths."
+          },
+          {
+            "left": "issue:43606",
+            "right": "issue:43819",
+            "accept": false,
+            "reason": "Different audio-model bugs: CPU offload device mismatch vs DAC latent/STE mismatch."
+          },
+          {
+            "left": "issue:43262",
+            "right": "issue:43746",
+            "accept": false,
+            "reason": "Different GraniteSpeech-related issues: audio chat-template sampling rate default vs PEFT adapter local checkpoint loading."
+          },
+          {
+            "left": "issue:41720",
+            "right": "issue:45823",
+            "accept": false,
+            "reason": "Different device-map failures on different models; no evidence of the same underlying bug."
+          },
+          {
+            "left": "issue:43526",
+            "right": "issue:43883",
+            "accept": false,
+            "reason": "Different regressions: BeitImageProcessorFast label reduction bug vs missing all_tied_weights_keys on Molmo."
+          },
+          {
+            "left": "issue:43454",
+            "right": "issue:43550",
+            "accept": false,
+            "reason": "Different bugs: tied lm_head weight issue vs torch.compile SDPA failure in Bamba-9B-v2."
+          },
+          {
+            "left": "issue:43377",
+            "right": "issue:44610",
+            "accept": false,
+            "reason": "Different problems: batching/padding-mask output mismatch in MIMI vs OmDet-Turbo processor/model input-size mismatch."
+          },
+          {
+            "left": "issue:43377",
+            "right": "issue:44898",
+            "accept": false,
+            "reason": "Different bugs in different models: MIMI batching mismatch vs Perceiver resolution handling."
+          },
+          {
+            "left": "issue:43454",
+            "right": "issue:45072",
+            "accept": false,
+            "reason": "Different bugs: lm_head weight tying vs dtype mismatch during bfloat16 inference."
+          },
+          {
+            "left": "issue:40990",
+            "right": "issue:45584",
+            "accept": false,
+            "reason": "Unrelated failures: perplexity anomaly on gpt-oss-20b vs Whisper empty-transcription generation failure."
+          },
+          {
+            "left": "issue:43572",
+            "right": "issue:45759",
+            "accept": false,
+            "reason": "Different issues: missing pad_token_idx in StableLmConfig vs from_config not unwrapping text_config for multimodal Qwen configs."
+          },
+          {
+            "left": "issue:39692",
+            "right": "issue:40444",
+            "accept": false,
+            "reason": "Different models and bugs: SigLIP2 documentation example errors vs Qwen2.5-VL multi-image finetuning failure."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 11,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:34634",
+        "issue:35707",
+        "issue:36331",
+        "issue:41720",
+        "issue:42617",
+        "issue:42915",
+        "issue:43296",
+        "issue:43316",
+        "issue:43317",
+        "issue:43322",
+        "issue:43404",
+        "issue:43450",
+        "issue:43454",
+        "issue:43526",
+        "issue:43550",
+        "issue:43577",
+        "issue:43606",
+        "issue:43708",
+        "issue:43720",
+        "issue:43749",
+        "issue:43792",
+        "issue:43873",
+        "issue:43881",
+        "issue:44008",
+        "issue:44186",
+        "issue:44387",
+        "issue:44423",
+        "issue:44466",
+        "issue:44479",
+        "issue:44534",
+        "issue:44610",
+        "issue:44683",
+        "issue:44756",
+        "issue:44877",
+        "issue:44898",
+        "issue:44913",
+        "issue:44918",
+        "issue:44977",
+        "issue:44991",
+        "issue:45072",
+        "issue:45357",
+        "issue:45406",
+        "issue:45632",
+        "issue:45684",
+        "issue:45701",
+        "issue:45704",
+        "issue:45823"
+      ],
+      "soft_pairs": [
+        "issue:43881|issue:44991",
+        "issue:44466|issue:45357",
+        "issue:43792|issue:44683",
+        "issue:43296|issue:44877",
+        "issue:43454|issue:43720",
+        "issue:43873|issue:45701",
+        "issue:41720|issue:44977",
+        "issue:44008|issue:45823",
+        "issue:45632|issue:45684",
+        "issue:43404|issue:45357",
+        "issue:34634|issue:35707",
+        "issue:44423|issue:45406",
+        "issue:44186|issue:45072",
+        "issue:43577|issue:43708",
+        "issue:35707|issue:41720",
+        "issue:43873|issue:44756",
+        "issue:44877|issue:44991",
+        "issue:43749|issue:44534",
+        "issue:42617|issue:42915",
+        "issue:43450|issue:45701",
+        "issue:43550|issue:43720",
+        "issue:36331|issue:42915",
+        "issue:44610|issue:44898",
+        "issue:44186|issue:44610",
+        "issue:36331|issue:42617",
+        "issue:44387|issue:45704",
+        "issue:44479|issue:44918",
+        "issue:43526|issue:43881",
+        "issue:42915|issue:43317",
+        "issue:43316|issue:44913",
+        "issue:43550|issue:43606",
+        "issue:41720|issue:43322"
+      ],
+      "prepared_review_unit_hash": "982c472f21e17977c55557050a8d624b5ca3568bc9689ba77260645181c4a213",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 47,
+        "item_count": 47,
+        "soft_pair_count": 32,
+        "serialized_chars": 36687,
+        "estimated_input_tokens": 9172,
+        "estimated_eval_tokens": 18600
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "This cluster is heterogeneous: the issues span unrelated bugs in loading, configs, generation, offloading, processors, and trainer behavior. I found no true duplicate pairs; all soft matches should be rejected.",
+        "confidence": 0.93,
+        "canonical_issue_reason": "No clear duplicate anchor exists. If a representative is required, issue 43873 is the broadest and most referenced offloading/quantization report, but it is not the same underlying bug as the others.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "Issue 43873 is the best representative by breadth and activity, but only as a loose thematic anchor; it should not be treated as a duplicate of the rest.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:43881",
+            "right": "issue:44991",
+            "accept": false,
+            "reason": "Different failure classes: model loading for glm-4v-9b vs tokenizer loading for a specific RoBERTa model."
+          },
+          {
+            "left": "issue:44466",
+            "right": "issue:45357",
+            "accept": false,
+            "reason": "Both involve serialization, but one is tied lm_head weights in v5 and the other is incorrect visual encoder keys for Qwen3.5."
+          },
+          {
+            "left": "issue:43792",
+            "right": "issue:44683",
+            "accept": false,
+            "reason": "Unrelated paths: Whisper model runtime failure vs compiled flex_attention breakage on newer torch."
+          },
+          {
+            "left": "issue:43296",
+            "right": "issue:44877",
+            "accept": false,
+            "reason": "Different loader issues: PaddleOCR-VL model loading vs strict config rejecting granite_speech config."
+          },
+          {
+            "left": "issue:43454",
+            "right": "issue:43720",
+            "accept": false,
+            "reason": "Both mention loading and weights, but they are different models and different bugs: tied lm_head weights vs unpacking BitNet packed weights."
+          },
+          {
+            "left": "issue:43873",
+            "right": "issue:45701",
+            "accept": false,
+            "reason": "Offloading/quantization behavior is unrelated to tokenizer changes across versions."
+          },
+          {
+            "left": "issue:41720",
+            "right": "issue:44977",
+            "accept": false,
+            "reason": "Same model family, but different code paths and symptoms: device-map CUDA assert vs flash-attention generation regression."
+          },
+          {
+            "left": "issue:44008",
+            "right": "issue:45823",
+            "accept": false,
+            "reason": "Different Gemma variants and different bugs: forward variable-name collision vs PLE device mismatch."
+          },
+          {
+            "left": "issue:45632",
+            "right": "issue:45684",
+            "accept": false,
+            "reason": "Both touch save/cache flows, but one is a trust_remote_code cache collision and the other is permission propagation in save_pretrained."
+          },
+          {
+            "left": "issue:43404",
+            "right": "issue:45357",
+            "accept": false,
+            "reason": "Both involve model serialization/weights, but one is Mistral3 tied weights and the other is Qwen3.5 visual encoder keys."
+          },
+          {
+            "left": "issue:34634",
+            "right": "issue:35707",
+            "accept": false,
+            "reason": "Completely different areas: Bark voice preset handling vs progressive generation with inputs_embeds/past_key_values."
+          },
+          {
+            "left": "issue:44423",
+            "right": "issue:45406",
+            "accept": false,
+            "reason": "Both are serve/processor crashes, but the concrete failures differ: multimodal .to on str vs Gemma4Processor missing _tokenizer."
+          },
+          {
+            "left": "issue:44186",
+            "right": "issue:45072",
+            "accept": false,
+            "reason": "Different subsystems and symptoms: tokenizer padding/NER crash vs dtype mismatch in inference wrappers."
+          },
+          {
+            "left": "issue:43577",
+            "right": "issue:43708",
+            "accept": false,
+            "reason": "BLIP2 dtype loading bug and Trainer max_steps resume math are unrelated."
+          },
+          {
+            "left": "issue:35707",
+            "right": "issue:41720",
+            "accept": false,
+            "reason": "Progressive generation bug and device-map CUDA assert are different failure modes with no shared code-path."
+          },
+          {
+            "left": "issue:43873",
+            "right": "issue:44756",
+            "accept": false,
+            "reason": "Both relate to memory/offloading, but one is quantization offload behavior and the other is mmap on Strix Halo causing OOM."
+          },
+          {
+            "left": "issue:44877",
+            "right": "issue:44991",
+            "accept": false,
+            "reason": "Config strictness blocking granite_speech is not the same as tokenizer loading failing on a specific model in v5."
+          },
+          {
+            "left": "issue:43749",
+            "right": "issue:44534",
+            "accept": false,
+            "reason": "Both concern model loading/state handling, but one is FSDP CPU RAM efficient loading and the other is non-persistent buffers being filled with junk."
+          },
+          {
+            "left": "issue:42617",
+            "right": "issue:42915",
+            "accept": false,
+            "reason": "Unrelated issues: 3d_parallel.py execution failure vs Qwen3Moe FineGrainedFP8Config failure."
+          },
+          {
+            "left": "issue:43450",
+            "right": "issue:45701",
+            "accept": false,
+            "reason": "Video processor batch-shape bug and tokenizer behavior changes are unrelated."
+          },
+          {
+            "left": "issue:43550",
+            "right": "issue:43720",
+            "accept": false,
+            "reason": "Different model/code paths: torch.compile SDPA failure in Bamba vs packed-weight unpacking in BitNet loading."
+          },
+          {
+            "left": "issue:36331",
+            "right": "issue:42915",
+            "accept": false,
+            "reason": "Trainer compute_loss signature mismatch is unrelated to Qwen3Moe FP8 loading."
+          },
+          {
+            "left": "issue:44610",
+            "right": "issue:44898",
+            "accept": false,
+            "reason": "Both involve image preprocessing size, but they are different models and different bugs, not the same underlying issue."
+          },
+          {
+            "left": "issue:44186",
+            "right": "issue:44610",
+            "accept": false,
+            "reason": "Tokenizer crash in LayoutLMv2 and OmDet-Turbo resolution mismatch are unrelated."
+          },
+          {
+            "left": "issue:36331",
+            "right": "issue:42617",
+            "accept": false,
+            "reason": "Trainer API mismatch and 3d_parallel execution failure are unrelated."
+          },
+          {
+            "left": "issue:44387",
+            "right": "issue:45704",
+            "accept": false,
+            "reason": "Both are memory-related, but one is int4 quantization OOM in Transformers 5.x and the other is a T5-apex RMSNorm memory leak."
+          },
+          {
+            "left": "issue:44479",
+            "right": "issue:44918",
+            "accept": false,
+            "reason": "Different multimodal failures: video-input regression in Qwen VL variants vs TRL SFT unpacking of input embeddings."
+          },
+          {
+            "left": "issue:43526",
+            "right": "issue:43881",
+            "accept": false,
+            "reason": "BEIT label reduction bug and glm-4v-9b load failure are unrelated."
+          },
+          {
+            "left": "issue:42915",
+            "right": "issue:43317",
+            "accept": false,
+            "reason": "Both are loading/config issues, but one is Qwen3Moe FP8 config failure and the other is dequantized model loading with gpu+cpu offload."
+          },
+          {
+            "left": "issue:43316",
+            "right": "issue:44913",
+            "accept": false,
+            "reason": "Both are config/API persistence issues, but the concrete fields and models differ: Gemma3TextConfig API discrepancy vs GPTNeoX rotary_pct reload."
+          },
+          {
+            "left": "issue:43550",
+            "right": "issue:43606",
+            "accept": false,
+            "reason": "Unrelated problems: Bamba SDPA torch.compile failure vs Bark-small CPU offload device mismatch."
+          },
+          {
+            "left": "issue:41720",
+            "right": "issue:43322",
+            "accept": false,
+            "reason": "Different bugs and symptoms: Qwen3 device-map CUDA assert vs Llava Next segmentation fault on load."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 12,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:30064",
+        "issue:34567",
+        "issue:34634",
+        "issue:35707",
+        "issue:36010",
+        "issue:40990",
+        "issue:41720",
+        "issue:42617",
+        "issue:42915",
+        "issue:43295",
+        "issue:43322",
+        "issue:43388",
+        "issue:43421",
+        "issue:43531",
+        "issue:43540",
+        "issue:43550",
+        "issue:43572",
+        "issue:43606",
+        "issue:43653",
+        "issue:43720",
+        "issue:43782",
+        "issue:43873",
+        "issue:43927",
+        "issue:44186",
+        "issue:44361",
+        "issue:44368",
+        "issue:44451",
+        "issue:44492",
+        "issue:44509",
+        "issue:44514",
+        "issue:44545",
+        "issue:44671",
+        "issue:44743",
+        "issue:44779",
+        "issue:44898",
+        "issue:44991",
+        "issue:45059",
+        "issue:45072",
+        "issue:45081",
+        "issue:45161",
+        "issue:45357",
+        "issue:45406",
+        "issue:45464",
+        "issue:45701",
+        "issue:45704",
+        "issue:45759",
+        "issue:45812",
+        "issue:45859"
+      ],
+      "soft_pairs": [
+        "issue:45081|issue:45812",
+        "issue:40990|issue:45701",
+        "issue:43295|issue:45701",
+        "issue:41720|issue:45161",
+        "issue:43531|issue:43572",
+        "issue:44671|issue:44743",
+        "issue:43421|issue:45812",
+        "issue:41720|issue:42617",
+        "issue:36010|issue:42617",
+        "issue:43540|issue:43782",
+        "issue:44186|issue:44361",
+        "issue:45406|issue:45464",
+        "issue:43421|issue:45701",
+        "issue:30064|issue:34634",
+        "issue:44492|issue:44509",
+        "issue:44779|issue:45812",
+        "issue:42915|issue:43322",
+        "issue:43606|issue:43720",
+        "issue:43388|issue:45059",
+        "issue:43550|issue:43653",
+        "issue:44514|issue:44545",
+        "issue:43531|issue:45759",
+        "issue:43606|issue:43653",
+        "issue:43421|issue:43927",
+        "issue:34567|issue:35707",
+        "issue:44368|issue:45357",
+        "issue:44451|issue:44991",
+        "issue:43873|issue:45704",
+        "issue:45759|issue:45859",
+        "issue:44898|issue:45072"
+      ],
+      "prepared_review_unit_hash": "2d3a809d1d86f0264bcbebdb2821585b30eab274778dae8d34caab27e99722b8",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 30,
+        "serialized_chars": 36745,
+        "estimated_input_tokens": 9187,
+        "estimated_eval_tokens": 18630
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "Only one soft pair looks like a true duplicate: the two Qwen2_5_VLProcessor.apply_chat_template crash reports. The rest are different models, code paths, or unrelated bug classes.",
+        "confidence": 0.95,
+        "canonical_issue_reason": "issue:44514 is the cleaner canonical issue: it was opened earlier and has the more complete report for the same batched-input `padding=False` crash.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "issue:44514 is the best issue overall for the same reason: it is the same bug as 44545 but has the fuller, earlier report.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:45081",
+            "right": "issue:45812",
+            "accept": false,
+            "reason": "Different tokenizer bugs for different model families (Mistral regex patch crash vs Granite token ID regression)."
+          },
+          {
+            "left": "issue:40990",
+            "right": "issue:45701",
+            "accept": false,
+            "reason": "Different symptoms and likely roots: perplexity evaluation problem vs tokenization behavior changes across versions."
+          },
+          {
+            "left": "issue:43295",
+            "right": "issue:45701",
+            "accept": false,
+            "reason": "Processor/tokenizer regression for multimodal custom code vs general tokenization change report; not the same bug."
+          },
+          {
+            "left": "issue:41720",
+            "right": "issue:45161",
+            "accept": false,
+            "reason": "Different model/runtime failures: CUDA assert with auto device mapping vs tensor-parallel not working on GPT-OSS MoE."
+          },
+          {
+            "left": "issue:43531",
+            "right": "issue:43572",
+            "accept": false,
+            "reason": "Unrelated issues in different components: Qwen3-MoE sliding window vs StableLmConfig pad_token_idx regression."
+          },
+          {
+            "left": "issue:44671",
+            "right": "issue:44743",
+            "accept": false,
+            "reason": "Different code paths and models: CamemBERT MLM outputs vs recurrent state reset in modular Qwen3.5 cache handling."
+          },
+          {
+            "left": "issue:43421",
+            "right": "issue:45812",
+            "accept": false,
+            "reason": "Runtime special-token post-processing update bug is distinct from Granite tokenizer wrong IDs."
+          },
+          {
+            "left": "issue:41720",
+            "right": "issue:42617",
+            "accept": false,
+            "reason": "Different failures in different features; no shared concrete bug path."
+          },
+          {
+            "left": "issue:36010",
+            "right": "issue:42617",
+            "accept": false,
+            "reason": "ImportError for GenerationMixin is unrelated to a 3d_parallel.py runtime failure."
+          },
+          {
+            "left": "issue:43540",
+            "right": "issue:43782",
+            "accept": false,
+            "reason": "Different multimodal model issues: video input processing vs from_pretrained weight_only loading error."
+          },
+          {
+            "left": "issue:44186",
+            "right": "issue:44361",
+            "accept": false,
+            "reason": "Different tokenizer/task crashes; same broad area but not the same underlying bug."
+          },
+          {
+            "left": "issue:45406",
+            "right": "issue:45464",
+            "accept": false,
+            "reason": "Gemma4Processor missing _tokenizer in serve is unrelated to Qwen3.5 streaming API failure."
+          },
+          {
+            "left": "issue:43421",
+            "right": "issue:45701",
+            "accept": false,
+            "reason": "TokenizersBackend post-processor sync bug is not the same as a general tokenization-version change report."
+          },
+          {
+            "left": "issue:30064",
+            "right": "issue:34634",
+            "accept": false,
+            "reason": "Completely different components and behaviors: image segmentation maps vs BarkProcessor voice preset."
+          },
+          {
+            "left": "issue:44492",
+            "right": "issue:44509",
+            "accept": false,
+            "reason": "Both are docs-related, but one is a typo and the other is removed pipeline-task documentation; not duplicates."
+          },
+          {
+            "left": "issue:44779",
+            "right": "issue:45812",
+            "accept": false,
+            "reason": "Different tokenizer regressions affecting different model families and outputs."
+          },
+          {
+            "left": "issue:42915",
+            "right": "issue:43322",
+            "accept": false,
+            "reason": "Different model/runtime failures: FP8 config issue vs segmentation fault on Llava Next loading."
+          },
+          {
+            "left": "issue:43606",
+            "right": "issue:43720",
+            "accept": false,
+            "reason": "Both involve loading/offload, but the concrete bugs are different model/code paths and not mergeable as one fix."
+          },
+          {
+            "left": "issue:43388",
+            "right": "issue:45059",
+            "accept": false,
+            "reason": "Different subsystems and symptoms: gather_for_metrics label truncation vs SAM3 PCS text/box behavior."
+          },
+          {
+            "left": "issue:43550",
+            "right": "issue:43653",
+            "accept": false,
+            "reason": "Different areas entirely: torch.compile/SDPA vs tokenizer special-token registration."
+          },
+          {
+            "left": "issue:44514",
+            "right": "issue:44545",
+            "accept": true,
+            "reason": "Same exact crash report: `Qwen2_5_VLProcessor.apply_chat_template` fails on batched input with `padding=False`."
+          },
+          {
+            "left": "issue:43531",
+            "right": "issue:45759",
+            "accept": false,
+            "reason": "Different Qwen model/config bugs: sliding window behavior vs `from_config` text_config unwrapping."
+          },
+          {
+            "left": "issue:43606",
+            "right": "issue:43653",
+            "accept": false,
+            "reason": "Different bug classes and model families; no shared underlying change."
+          },
+          {
+            "left": "issue:43421",
+            "right": "issue:43927",
+            "accept": false,
+            "reason": "Runtime special-token update bug is unrelated to DiaConfig save/load token ID loss."
+          },
+          {
+            "left": "issue:34567",
+            "right": "issue:35707",
+            "accept": false,
+            "reason": "TrainerState token count tracking is unrelated to progressive generation with embeds and cache."
+          },
+          {
+            "left": "issue:44368",
+            "right": "issue:45357",
+            "accept": false,
+            "reason": "Different Qwen3.5 issues: config warning vs incorrect visual encoder keys on save_pretrained."
+          },
+          {
+            "left": "issue:44451",
+            "right": "issue:44991",
+            "accept": false,
+            "reason": "Both are tokenizer-loading failures, but for different models and with no evidence of the same root cause."
+          },
+          {
+            "left": "issue:43873",
+            "right": "issue:45704",
+            "accept": false,
+            "reason": "Quantization/offload behavior is unrelated to the apex FusedRMSNorm memory leak."
+          },
+          {
+            "left": "issue:45759",
+            "right": "issue:45859",
+            "accept": false,
+            "reason": "Different Qwen 3.5/3.6 multimodal config handling vs CPU-offload incompatibility in rotary embedding."
+          },
+          {
+            "left": "issue:44898",
+            "right": "issue:45072",
+            "accept": false,
+            "reason": "Different model execution problems: Perceiver positional interpolation vs dtype mismatch in bfloat16 inference."
+          }
+        ]
+      },
+      "evaluator_result": {
+        "accept": true,
+        "feedback": "The conclusions are grounded in the packet. The lone accepted pair, issue:44514/issue:44545, is an exact title match and the earlier issue also has the fuller report, so the canonicalization choice is conservative. The other soft-edge rejections stay within what the titles support and avoid overstating duplicate evidence."
+      },
+      "evaluator_used": true,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": true,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 13,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:30064",
+        "issue:33666",
+        "issue:34634",
+        "issue:35707",
+        "issue:36010",
+        "issue:39692",
+        "issue:41720",
+        "issue:42886",
+        "issue:42915",
+        "issue:43381",
+        "issue:43421",
+        "issue:43525",
+        "issue:43531",
+        "issue:43540",
+        "issue:43596",
+        "issue:43618",
+        "issue:43653",
+        "issue:43697",
+        "issue:43746",
+        "issue:43782",
+        "issue:43844",
+        "issue:43873",
+        "issue:43994",
+        "issue:44008",
+        "issue:44062",
+        "issue:44186",
+        "issue:44220",
+        "issue:44451",
+        "issue:44589",
+        "issue:44610",
+        "issue:44625",
+        "issue:44661",
+        "issue:44671",
+        "issue:44843",
+        "issue:44918",
+        "issue:44936",
+        "issue:44987",
+        "issue:45042",
+        "issue:45081",
+        "issue:45183",
+        "issue:45356",
+        "issue:45381",
+        "issue:45479",
+        "issue:45588",
+        "issue:45704",
+        "issue:45759",
+        "issue:45859"
+      ],
+      "soft_pairs": [
+        "issue:41720|issue:43746",
+        "issue:44062|issue:44451",
+        "issue:44987|issue:45042",
+        "issue:43531|issue:44589",
+        "issue:33666|issue:35707",
+        "issue:43421|issue:43653",
+        "issue:45588|issue:45759",
+        "issue:42886|issue:44843",
+        "issue:43381|issue:43844",
+        "issue:44918|issue:44936",
+        "issue:39692|issue:44610",
+        "issue:43421|issue:44186",
+        "issue:43525|issue:43531",
+        "issue:44625|issue:45479",
+        "issue:45704|issue:45859",
+        "issue:43782|issue:45381",
+        "issue:45081|issue:45356",
+        "issue:34634|issue:41720",
+        "issue:43994|issue:44671",
+        "issue:43381|issue:45704",
+        "issue:39692|issue:43873",
+        "issue:42915|issue:45759",
+        "issue:43596|issue:44220",
+        "issue:43540|issue:44008",
+        "issue:43618|issue:43697",
+        "issue:44220|issue:44661",
+        "issue:45183|issue:45479",
+        "issue:30064|issue:36010"
+      ],
+      "prepared_review_unit_hash": "cbafb5631f9b38104abeec8db3b4f4af8f7eaff0dcff30a64587688523acc641",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 47,
+        "item_count": 47,
+        "soft_pair_count": 28,
+        "serialized_chars": 35665,
+        "estimated_input_tokens": 8917,
+        "estimated_eval_tokens": 18090
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "This cluster is highly heterogeneous. The soft links are mostly broad semantic overlaps (tokenizers, Qwen, gradient/checkpointing, image processors) rather than the same concrete bug, so they should be rejected. No PRs are present.",
+        "confidence": 0.79,
+        "canonical_issue_reason": "issue:44843 is the most concrete, broadly relevant issue in the cluster and has a clear, actionable tokenizer/offline failure mode; it is also the closest to another soft pair in this set.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "issue:44843 is a good representative because it describes a specific reproducible regression with a clearly identified code path (_patch_mistral_regex / HF_HUB_OFFLINE) and is more general than the other similarly themed tokenizer issues.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:41720",
+            "right": "issue:43746",
+            "accept": false,
+            "reason": "Different models and failures: Qwen3 device-mapping CUDA assert vs GraniteSpeech PEFT checkpoint loading."
+          },
+          {
+            "left": "issue:44062",
+            "right": "issue:44451",
+            "accept": false,
+            "reason": "Different bugs: AddedToken special-kwarg conflict vs failure to load a specific model checkpoint."
+          },
+          {
+            "left": "issue:44987",
+            "right": "issue:45042",
+            "accept": false,
+            "reason": "Unrelated failures: model loading for a specific repo vs PIL image processors requiring torchvision."
+          },
+          {
+            "left": "issue:43531",
+            "right": "issue:44589",
+            "accept": false,
+            "reason": "Different code paths: Qwen3-MoE sliding_window behavior vs missing Float8 storage type."
+          },
+          {
+            "left": "issue:33666",
+            "right": "issue:35707",
+            "accept": false,
+            "reason": "Qwen2-VL multi-GPU training is unrelated to progressive generation with inputs_embeds and past_key_values."
+          },
+          {
+            "left": "issue:43421",
+            "right": "issue:43653",
+            "accept": false,
+            "reason": "Both involve special tokens, but one is runtime post-processor updates and the other is BigBirdTokenizer mask-token registration."
+          },
+          {
+            "left": "issue:45588",
+            "right": "issue:45759",
+            "accept": false,
+            "reason": "Different concrete problems: flash_attention s_aux=None crash vs missing text_config unwrap in composite Qwen configs."
+          },
+          {
+            "left": "issue:42886",
+            "right": "issue:44843",
+            "accept": false,
+            "reason": "Both concern offline behavior, but one is cache loading and the other is an unconditional model_info call in _patch_mistral_regex."
+          },
+          {
+            "left": "issue:43381",
+            "right": "issue:43844",
+            "accept": false,
+            "reason": "Both mention gradients, but one is eval-mode checkpointing misuse and the other is ZeRO-3 gradient explosion."
+          },
+          {
+            "left": "issue:44918",
+            "right": "issue:44936",
+            "accept": false,
+            "reason": "Trainer/SFT issues, but they are distinct failures: unpacking embeddings vs evaluate-after-train behavior."
+          },
+          {
+            "left": "issue:39692",
+            "right": "issue:44610",
+            "accept": false,
+            "reason": "Both are model/processor mismatches, but they affect different models and different preprocessing mistakes."
+          },
+          {
+            "left": "issue:43421",
+            "right": "issue:44186",
+            "accept": false,
+            "reason": "Different tokenizer defects: runtime post-processor sync vs NER/batched padding crashes."
+          },
+          {
+            "left": "issue:43525",
+            "right": "issue:43531",
+            "accept": false,
+            "reason": "Unrelated Qwen3-MoE sliding_window bug vs missing pad_token_id on Llama4Config."
+          },
+          {
+            "left": "issue:44625",
+            "right": "issue:45479",
+            "accept": false,
+            "reason": "Both involve num_labels, but one is config propagation and the other is degenerate loss for num_labels=1."
+          },
+          {
+            "left": "issue:45704",
+            "right": "issue:45859",
+            "accept": false,
+            "reason": "Different components: T5 apex RMSNorm memory leak vs CPU-offload incompatibility in Qwen3_5MoeTextRotaryEmbedding."
+          },
+          {
+            "left": "issue:43782",
+            "right": "issue:45381",
+            "accept": false,
+            "reason": "Different multimodal bugs: weight_only loading error vs wrong video vision_position_ids."
+          },
+          {
+            "left": "issue:45081",
+            "right": "issue:45356",
+            "accept": false,
+            "reason": "Related tokenizer area, but not the same bug: one crashes in _patch_mistral_regex, the other is a Kimi-K2.5 regression with codec handling/warning text."
+          },
+          {
+            "left": "issue:34634",
+            "right": "issue:41720",
+            "accept": false,
+            "reason": "Different models and behaviors: BarkProcessor voice_preset vs Qwen3 auto device mapping CUDA assert."
+          },
+          {
+            "left": "issue:43994",
+            "right": "issue:44671",
+            "accept": false,
+            "reason": "Different model-specific output regressions: SigLIP2 pipeline/model mismatch vs CamemBERT masked LM predictions."
+          },
+          {
+            "left": "issue:43381",
+            "right": "issue:45704",
+            "accept": false,
+            "reason": "Unrelated gradient-checkpointing mode issue vs T5 RMSNorm memory leak."
+          },
+          {
+            "left": "issue:39692",
+            "right": "issue:43873",
+            "accept": false,
+            "reason": "Different problems: documentation/model mismatch vs quantization offloading behavior."
+          },
+          {
+            "left": "issue:42915",
+            "right": "issue:45759",
+            "accept": false,
+            "reason": "Different Qwen families and failures: FineGrainedFP8Config error vs missing text_config unwrap."
+          },
+          {
+            "left": "issue:43596",
+            "right": "issue:44220",
+            "accept": false,
+            "reason": "Different runtime errors in unrelated paths: ZeRO-3/BertModel init vs _torch_extract_fbank_features."
+          },
+          {
+            "left": "issue:43540",
+            "right": "issue:44008",
+            "accept": false,
+            "reason": "Different multimodal failures: Qwen3OmniMoe video processing vs Gemma 3n audio_mel_mask name collision."
+          },
+          {
+            "left": "issue:43618",
+            "right": "issue:43697",
+            "accept": false,
+            "reason": "Different issues: CLIPOutput attentions assignment regression vs RTDetrV2 output differences in v5."
+          },
+          {
+            "left": "issue:44220",
+            "right": "issue:44661",
+            "accept": false,
+            "reason": "Different subsystems: audio feature extraction vs add-new-model-like/tokenizer mapping logic."
+          },
+          {
+            "left": "issue:45183",
+            "right": "issue:45479",
+            "accept": false,
+            "reason": "Generic validation error messaging vs a sequence-classification loss/label bug; not the same underlying defect."
+          },
+          {
+            "left": "issue:30064",
+            "right": "issue:36010",
+            "accept": false,
+            "reason": "Completely unrelated: void segmentation map processing vs GenerationMixin import failure."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 14,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:30064",
+        "issue:36331",
+        "issue:42915",
+        "issue:43295",
+        "issue:43322",
+        "issue:43335",
+        "issue:43421",
+        "issue:43531",
+        "issue:43540",
+        "issue:43550",
+        "issue:43577",
+        "issue:43643",
+        "issue:43653",
+        "issue:43716",
+        "issue:43720",
+        "issue:43746",
+        "issue:43756",
+        "issue:43782",
+        "issue:43818",
+        "issue:43819",
+        "issue:43825",
+        "issue:43873",
+        "issue:43994",
+        "issue:44008",
+        "issue:44062",
+        "issue:44077",
+        "issue:44186",
+        "issue:44448",
+        "issue:44485",
+        "issue:44488",
+        "issue:44509",
+        "issue:44561",
+        "issue:44625",
+        "issue:44671",
+        "issue:44811",
+        "issue:44821",
+        "issue:44843",
+        "issue:44877",
+        "issue:44898",
+        "issue:44977",
+        "issue:45020",
+        "issue:45072",
+        "issue:45081",
+        "issue:45084",
+        "issue:45381",
+        "issue:45584",
+        "issue:45685",
+        "issue:45834"
+      ],
+      "soft_pairs": [
+        "issue:43720|issue:43819",
+        "issue:43335|issue:43756",
+        "issue:44077|issue:44625",
+        "issue:43653|issue:43720",
+        "issue:43716|issue:44008",
+        "issue:42915|issue:43746",
+        "issue:43295|issue:45834",
+        "issue:43746|issue:45381",
+        "issue:43550|issue:45072",
+        "issue:43746|issue:43873",
+        "issue:43550|issue:43819",
+        "issue:44186|issue:44898",
+        "issue:44843|issue:45081",
+        "issue:44062|issue:44488",
+        "issue:43531|issue:43643",
+        "issue:43577|issue:43818",
+        "issue:43643|issue:44877",
+        "issue:43421|issue:44811",
+        "issue:43540|issue:45381",
+        "issue:43653|issue:43819",
+        "issue:44448|issue:44671",
+        "issue:43782|issue:44977",
+        "issue:45584|issue:45834",
+        "issue:43322|issue:45685",
+        "issue:43825|issue:43994",
+        "issue:44561|issue:45020",
+        "issue:44821|issue:45084",
+        "issue:44485|issue:44509",
+        "issue:30064|issue:36331"
+      ],
+      "prepared_review_unit_hash": "581afd0ed426f7995d3c02bdaaa8d01fedf7bd5c76d40daf4317db52df2d72d9",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 29,
+        "serialized_chars": 36556,
+        "estimated_input_tokens": 9139,
+        "estimated_eval_tokens": 18534
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "This set is highly heterogeneous: most items are unrelated model-, tokenizer-, config-, or docs-level regressions. The soft matches mostly share only subsystem vocabulary, not the same concrete bug, so none should be merged as duplicates.",
+        "confidence": 0.94,
+        "canonical_issue_reason": "If one issue must anchor the cluster, issue:45020 is the broadest/open report and best matches a general umbrella of recent remote_code regressions.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "Issue:45020 is the most representative of the cluster's broad regression theme, but only weakly so because the surrounding items are not true duplicates.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:43720",
+            "right": "issue:43819",
+            "accept": false,
+            "reason": "Both are loading/runtime failures, but one is BitNet packed-weight unpacking during accelerate loading and the other is DAC from_latents vs forward mismatch from missing STE. Different code paths and bugs."
+          },
+          {
+            "left": "issue:43335",
+            "right": "issue:43756",
+            "accept": false,
+            "reason": "SwitchTransformers config layer-count bug and Smollm3 RoPE-layer omission are unrelated model/config issues."
+          },
+          {
+            "left": "issue:44077",
+            "right": "issue:44625",
+            "accept": false,
+            "reason": "One is about patchtsmixer post_init policy; the other is Qwen3.5 num_labels propagation. Same broad config area, different bugs."
+          },
+          {
+            "left": "issue:43653",
+            "right": "issue:43720",
+            "accept": false,
+            "reason": "Tokenizer special-token decode issue vs BitNet packed-weight loading bug; no shared underlying failure."
+          },
+          {
+            "left": "issue:43716",
+            "right": "issue:44008",
+            "accept": false,
+            "reason": "Both are multimodal dtype-ish reports, but one is Mistral image preprocessor/model dtype mismatch and the other is a Gemma3n variable-name collision in forward()."
+          },
+          {
+            "left": "issue:42915",
+            "right": "issue:43746",
+            "accept": false,
+            "reason": "Qwen3Moe FineGrainedFP8Config failure and GraniteSpeech PEFT checkpoint loading are unrelated."
+          },
+          {
+            "left": "issue:43295",
+            "right": "issue:45834",
+            "accept": false,
+            "reason": "Processor.tokenizer regression and Kosmos2.5 OCR long-input index error are different code paths."
+          },
+          {
+            "left": "issue:43746",
+            "right": "issue:45381",
+            "accept": false,
+            "reason": "GraniteSpeech local-checkpoint loading and qwen2.5-vl video position-id bug are unrelated model-specific issues."
+          },
+          {
+            "left": "issue:43550",
+            "right": "issue:45072",
+            "accept": false,
+            "reason": "Both mention dtype/compile concerns, but they affect different models and failures (Bamba torch.compile SDPA vs SwitchTransformers/TimmWrapper bfloat16 mismatch)."
+          },
+          {
+            "left": "issue:43746",
+            "right": "issue:43873",
+            "accept": false,
+            "reason": "GraniteSpeech PEFT loading and quantization offloading behavior are not the same bug."
+          },
+          {
+            "left": "issue:43550",
+            "right": "issue:43819",
+            "accept": false,
+            "reason": "Bamba SDPA torch.compile failure and DAC latent/STE mismatch are unrelated."
+          },
+          {
+            "left": "issue:44186",
+            "right": "issue:44898",
+            "accept": false,
+            "reason": "LayoutLMv2Tokenizer NER/padding crash and Perceiver interpolation failure are different tokenizer vs vision-model bugs."
+          },
+          {
+            "left": "issue:44843",
+            "right": "issue:45081",
+            "accept": false,
+            "reason": "Both touch _patch_mistral_regex, but one is an offline model_info call and the other is a backend_tokenizer AttributeError. Same helper, different failure mode."
+          },
+          {
+            "left": "issue:44062",
+            "right": "issue:44488",
+            "accept": false,
+            "reason": "AddedToken special-argument collision is unrelated to the sleng-bert loading problem."
+          },
+          {
+            "left": "issue:43531",
+            "right": "issue:43643",
+            "accept": false,
+            "reason": "Qwen3-MoE sliding_window behavior and trust_remote_code missing-fields config loading are unrelated."
+          },
+          {
+            "left": "issue:43577",
+            "right": "issue:43818",
+            "accept": false,
+            "reason": "Blip2 dtype persistence and Video-LLaVA temporal-attention/weight-sharing are different bugs."
+          },
+          {
+            "left": "issue:43643",
+            "right": "issue:44877",
+            "accept": false,
+            "reason": "Both are config-loading regressions, but one is trust_remote_code missing fields and the other is strict config rejection for granite_speech."
+          },
+          {
+            "left": "issue:43421",
+            "right": "issue:44811",
+            "accept": false,
+            "reason": "Special-token post-processor updates and Whisper skip_special_tokens handling are related only at a high level, not the same bug."
+          },
+          {
+            "left": "issue:43540",
+            "right": "issue:45381",
+            "accept": false,
+            "reason": "Both involve video inputs, but Qwen3OmniMoe video processing and qwen2.5-vl vision_position_ids are distinct issues."
+          },
+          {
+            "left": "issue:43653",
+            "right": "issue:43819",
+            "accept": false,
+            "reason": "BigBirdTokenizer mask-token decoding and DAC forward/from_latents mismatch are unrelated."
+          },
+          {
+            "left": "issue:44448",
+            "right": "issue:44671",
+            "accept": false,
+            "reason": "Both are v4/v5 behavior regressions, but on different models and code paths; not duplicates."
+          },
+          {
+            "left": "issue:43782",
+            "right": "issue:44977",
+            "accept": false,
+            "reason": "Qwen3VL weight_only loading error and Qwen3.5 flash-attention generation bug are different Qwen model failures."
+          },
+          {
+            "left": "issue:45584",
+            "right": "issue:45834",
+            "accept": false,
+            "reason": "Whisper empty-transcription generation and Kosmos2.5 long OCR index error are unrelated."
+          },
+          {
+            "left": "issue:43322",
+            "right": "issue:45685",
+            "accept": false,
+            "reason": "Llava Next segfault and MPS histogram Int error are unrelated runtime failures."
+          },
+          {
+            "left": "issue:43825",
+            "right": "issue:43994",
+            "accept": false,
+            "reason": "Docs/error-message mismatch for removed pipeline tasks and SigLIP2 bad outputs are different issues."
+          },
+          {
+            "left": "issue:44561",
+            "right": "issue:45020",
+            "accept": false,
+            "reason": "Both concern remote_code regressions, but 44561 is a specific missing utility and 45020 is a broad umbrella report; not the same underlying bug."
+          },
+          {
+            "left": "issue:44821",
+            "right": "issue:45084",
+            "accept": false,
+            "reason": "AutoImageProcessor URL loading and template-node compilation errors are unrelated."
+          },
+          {
+            "left": "issue:44485",
+            "right": "issue:44509",
+            "accept": false,
+            "reason": "GLM-5 RoPE implementation discussion and v5 docs cleanup are unrelated."
+          },
+          {
+            "left": "issue:30064",
+            "right": "issue:36331",
+            "accept": false,
+            "reason": "Void segmentation-map processing and CustomTrainer.compute_loss signature breakage have no common code path."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 15,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:34634",
+        "issue:36010",
+        "issue:39692",
+        "issue:42175",
+        "issue:43262",
+        "issue:43329",
+        "issue:43404",
+        "issue:43452",
+        "issue:43526",
+        "issue:43577",
+        "issue:43582",
+        "issue:43606",
+        "issue:43638",
+        "issue:43644",
+        "issue:43650",
+        "issue:43653",
+        "issue:43697",
+        "issue:43720",
+        "issue:43742",
+        "issue:43749",
+        "issue:43782",
+        "issue:43819",
+        "issue:43825",
+        "issue:43827",
+        "issue:43906",
+        "issue:44060",
+        "issue:44186",
+        "issue:44368",
+        "issue:44442",
+        "issue:44479",
+        "issue:44610",
+        "issue:44671",
+        "issue:44805",
+        "issue:44857",
+        "issue:44912",
+        "issue:44936",
+        "issue:44991",
+        "issue:45020",
+        "issue:45072",
+        "issue:45081",
+        "issue:45084",
+        "issue:45406",
+        "issue:45459",
+        "issue:45538",
+        "issue:45600",
+        "issue:45701",
+        "issue:45803",
+        "issue:45834"
+      ],
+      "soft_pairs": [
+        "issue:39692|issue:45834",
+        "issue:44479|issue:45834",
+        "issue:43825|issue:43827",
+        "issue:44186|issue:44442",
+        "issue:43606|issue:45072",
+        "issue:43577|issue:43582",
+        "issue:34634|issue:36010",
+        "issue:44805|issue:44936",
+        "issue:43452|issue:44991",
+        "issue:42175|issue:45081",
+        "issue:44610|issue:45072",
+        "issue:43720|issue:45072",
+        "issue:45081|issue:45406",
+        "issue:44060|issue:44368",
+        "issue:43650|issue:43906",
+        "issue:43404|issue:43782",
+        "issue:44671|issue:45701",
+        "issue:43582|issue:44857",
+        "issue:43262|issue:43819",
+        "issue:43526|issue:43697",
+        "issue:45459|issue:45803",
+        "issue:45084|issue:45538",
+        "issue:43644|issue:43749",
+        "issue:43329|issue:45600",
+        "issue:43653|issue:44912",
+        "issue:43742|issue:45020",
+        "issue:43638|issue:44936"
+      ],
+      "prepared_review_unit_hash": "cf251d1549d265c23d29f66c2da4d404863d954f73873b874f1cfd1dd8e15ae9",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 27,
+        "serialized_chars": 36120,
+        "estimated_input_tokens": 9030,
+        "estimated_eval_tokens": 18316
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "No soft edge looks like a true duplicate; the pairs are mostly unrelated model-specific regressions or, at most, adjacent documentation/runtime issues. The broadest issue in the set is the open remote_code regression tracker.",
+        "confidence": 0.7,
+        "canonical_issue_reason": "issue:45020 is the broadest open umbrella here, covering recent-version regressions that break model loading under remote_code; most other issues are narrower, model-specific bugs.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "issue:45020 best represents the cluster because it is open, general, and captures the shared Transformers-v5 regression theme better than the single-model reports.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:39692",
+            "right": "issue:45834",
+            "accept": false,
+            "reason": "SigLIP2 doc example errors and Kosmos2.5 OCR index errors are unrelated models and failure modes."
+          },
+          {
+            "left": "issue:44479",
+            "right": "issue:45834",
+            "accept": false,
+            "reason": "Qwen VLM video-input regression is a different code path from Kosmos2.5 long-OCR indexing."
+          },
+          {
+            "left": "issue:43825",
+            "right": "issue:43827",
+            "accept": false,
+            "reason": "One is a runtime error message about translation pipelines; the other is documentation still mentioning pipeline()."
+          },
+          {
+            "left": "issue:44186",
+            "right": "issue:44442",
+            "accept": false,
+            "reason": "LayoutLMv2 tokenizer NER/padding crashes and FastSpeech2Conformer tokenizer loading are unrelated tokenizer bugs."
+          },
+          {
+            "left": "issue:43606",
+            "right": "issue:45072",
+            "accept": false,
+            "reason": "Bark CPU offload device mismatch and SwitchTransformers/TimmWrapperModel dtype mismatch are different loading failures."
+          },
+          {
+            "left": "issue:43577",
+            "right": "issue:43582",
+            "accept": false,
+            "reason": "BLIP2 dtype propagation and Apple Silicon allocator warmup TypeError do not share the same bug."
+          },
+          {
+            "left": "issue:34634",
+            "right": "issue:36010",
+            "accept": false,
+            "reason": "BarkProcessor voice_preset behavior and GenerationMixin import failure are unrelated."
+          },
+          {
+            "left": "issue:44805",
+            "right": "issue:44936",
+            "accept": false,
+            "reason": "A tensor mask shape IndexError is not the same failure as trainer.evaluate() after trainer.train()."
+          },
+          {
+            "left": "issue:43452",
+            "right": "issue:44991",
+            "accept": false,
+            "reason": "gguf_file/pretrained-loading breakage and est-roberta tokenizer loading are different loader paths."
+          },
+          {
+            "left": "issue:42175",
+            "right": "issue:45081",
+            "accept": false,
+            "reason": "TensorFlow backend packaging and Mistral regex patch crashes are unrelated."
+          },
+          {
+            "left": "issue:44610",
+            "right": "issue:45072",
+            "accept": false,
+            "reason": "OmDet-Turbo image-size mismatch is unrelated to dtype mismatch inference bugs."
+          },
+          {
+            "left": "issue:43720",
+            "right": "issue:45072",
+            "accept": false,
+            "reason": "BitNet packed-weight unpacking and SwitchTransformers/TimmWrapperModel dtype issues are different concrete bugs."
+          },
+          {
+            "left": "issue:45081",
+            "right": "issue:45406",
+            "accept": false,
+            "reason": "Mistral tokenizer regex patching and Gemma4Processor missing _tokenizer are separate tokenizer/processor failures."
+          },
+          {
+            "left": "issue:44060",
+            "right": "issue:44368",
+            "accept": false,
+            "reason": "Both mention tied-weight warnings, but they concern different models and different underlying causes."
+          },
+          {
+            "left": "issue:43650",
+            "right": "issue:43906",
+            "accept": false,
+            "reason": "A placeholder 'ADD THE DATA' issue is not a duplicate of an isolated reproduction report."
+          },
+          {
+            "left": "issue:43404",
+            "right": "issue:43782",
+            "accept": false,
+            "reason": "Mistral lm_head tying and Qwen3VL weight_only loading error are unrelated model-specific problems."
+          },
+          {
+            "left": "issue:44671",
+            "right": "issue:45701",
+            "accept": false,
+            "reason": "CamemBERT masked-LM regression and a general tokenization-change report are not the same bug."
+          },
+          {
+            "left": "issue:43582",
+            "right": "issue:44857",
+            "accept": false,
+            "reason": "Apple Silicon allocator warmup TypeError and CUDA float16 loss crash are unrelated platform-specific issues."
+          },
+          {
+            "left": "issue:43262",
+            "right": "issue:43819",
+            "accept": false,
+            "reason": "Audio apply_chat_template sampling-rate defaulting and DAC latent reconstruction mismatch are different behaviors."
+          },
+          {
+            "left": "issue:43526",
+            "right": "issue:43697",
+            "accept": false,
+            "reason": "BEiT FastImageProcessor label reduction and RTDetrV2 output drift are unrelated model/processor bugs."
+          },
+          {
+            "left": "issue:45459",
+            "right": "issue:45803",
+            "accept": false,
+            "reason": "Both involve exception handling, but they are different functions with different failure surfaces."
+          },
+          {
+            "left": "issue:45084",
+            "right": "issue:45538",
+            "accept": false,
+            "reason": "Template-compilation TypeError and CLIPTokenizer model_max_length are unrelated tokenizer/template issues."
+          },
+          {
+            "left": "issue:43644",
+            "right": "issue:43749",
+            "accept": false,
+            "reason": "Non-persistent buffer junk and FSDP CPU RAM efficient loading breakage are different loading bugs."
+          },
+          {
+            "left": "issue:43329",
+            "right": "issue:45600",
+            "accept": false,
+            "reason": "Undefined video-token helpers and stale auto_mappings references are unrelated regressions."
+          },
+          {
+            "left": "issue:43653",
+            "right": "issue:44912",
+            "accept": false,
+            "reason": "BigBird special-token decode failure and MXFP4 quantization fallback are unrelated."
+          },
+          {
+            "left": "issue:43742",
+            "right": "issue:45020",
+            "accept": false,
+            "reason": "A key error for one model is too specific to be the same underlying bug as the broad remote_code regression tracker."
+          },
+          {
+            "left": "issue:43638",
+            "right": "issue:44936",
+            "accept": false,
+            "reason": "DeepSpeed ZeRO-3 with a non-pretrained Bert model and post-train evaluate failure are different training/runtime issues."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 16,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:34567",
+        "issue:34634",
+        "issue:36010",
+        "issue:36331",
+        "issue:38175",
+        "issue:39692",
+        "issue:40990",
+        "issue:41720",
+        "issue:43257",
+        "issue:43262",
+        "issue:43329",
+        "issue:43450",
+        "issue:43454",
+        "issue:43526",
+        "issue:43531",
+        "issue:43540",
+        "issue:43550",
+        "issue:43577",
+        "issue:43606",
+        "issue:43643",
+        "issue:43645",
+        "issue:43697",
+        "issue:43720",
+        "issue:43746",
+        "issue:43749",
+        "issue:43819",
+        "issue:43854",
+        "issue:43927",
+        "issue:43931",
+        "issue:44291",
+        "issue:44479",
+        "issue:44554",
+        "issue:44568",
+        "issue:44589",
+        "issue:44610",
+        "issue:44811",
+        "issue:44849",
+        "issue:44912",
+        "issue:44977",
+        "issue:45042",
+        "issue:45356",
+        "issue:45357",
+        "issue:45488",
+        "issue:45701",
+        "issue:45759",
+        "issue:45823",
+        "issue:45859"
+      ],
+      "soft_pairs": [
+        "issue:43746|issue:44811",
+        "issue:44589|issue:45042",
+        "issue:43854|issue:44977",
+        "issue:38175|issue:44610",
+        "issue:43526|issue:44554",
+        "issue:39692|issue:44912",
+        "issue:36010|issue:43645",
+        "issue:34634|issue:36331",
+        "issue:43819|issue:43927",
+        "issue:43257|issue:43329",
+        "issue:43454|issue:43577",
+        "issue:43720|issue:44610",
+        "issue:43550|issue:44912",
+        "issue:44568|issue:45488",
+        "issue:43749|issue:44291",
+        "issue:43531|issue:45042",
+        "issue:43262|issue:44811",
+        "issue:44568|issue:45356",
+        "issue:36010|issue:36331",
+        "issue:43643|issue:45759",
+        "issue:44849|issue:45701",
+        "issue:34567|issue:40990",
+        "issue:43329|issue:43450",
+        "issue:43329|issue:44479",
+        "issue:43606|issue:45859",
+        "issue:43257|issue:43931",
+        "issue:43697|issue:45357",
+        "issue:45823|issue:45859",
+        "issue:44291|issue:45042",
+        "issue:41720|issue:43540"
+      ],
+      "prepared_review_unit_hash": "34a231f632c19b9b90b7b070c40cdb5bfa1932ccfe4a7e7aa5eebc9620f3db25",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 47,
+        "item_count": 47,
+        "soft_pair_count": 30,
+        "serialized_chars": 36446,
+        "estimated_input_tokens": 9112,
+        "estimated_eval_tokens": 18480
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "The cluster is heterogeneous: it mixes unrelated Transformers regressions across loading, tokenization, vision/audio processors, and model-specific inference paths. None of the soft pairs looks like a true duplicate; the best representative is the broad tokenizer-regression issue.",
+        "confidence": 0.31,
+        "canonical_issue_reason": "issue:45701 is the broadest, most umbrella-style bug here and best matches the recurring tokenizer/version-regression theme; it is more representative than the many narrow model-specific reports.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "issue:45701 is the best cluster representative because it describes a version-wide tokenization regression rather than a single model or narrow code path, making it the most reusable anchor for this set.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:43746",
+            "right": "issue:44811",
+            "accept": false,
+            "reason": "Different subsystems: PEFT/local checkpoint loading vs Whisper batch decoding behavior."
+          },
+          {
+            "left": "issue:44589",
+            "right": "issue:45042",
+            "accept": false,
+            "reason": "Unrelated bugs: Float8 storage lookup vs PIL backend image-processor dependency handling."
+          },
+          {
+            "left": "issue:43854",
+            "right": "issue:44977",
+            "accept": false,
+            "reason": "Both are model-specific regressions, but they affect different models and different code paths (loading vs flash-attention generation)."
+          },
+          {
+            "left": "issue:38175",
+            "right": "issue:44610",
+            "accept": false,
+            "reason": "Vision-related, but one is a probability/output issue and the other is an input-size preprocessing mismatch."
+          },
+          {
+            "left": "issue:43526",
+            "right": "issue:44554",
+            "accept": false,
+            "reason": "Different areas: Beit image label reduction vs MPS attention correctness."
+          },
+          {
+            "left": "issue:39692",
+            "right": "issue:44912",
+            "accept": false,
+            "reason": "Docs/example issues and quantization fallback are not the same underlying bug."
+          },
+          {
+            "left": "issue:36010",
+            "right": "issue:43645",
+            "accept": false,
+            "reason": "Both involve loading/import behavior, but one is an import path break and the other is custom model initialization in notebooks."
+          },
+          {
+            "left": "issue:34634",
+            "right": "issue:36331",
+            "accept": false,
+            "reason": "Audio processor preset handling and Trainer API incompatibility are separate bugs."
+          },
+          {
+            "left": "issue:43819",
+            "right": "issue:43927",
+            "accept": false,
+            "reason": "Different code paths: DAC latent decoding vs config save/load token IDs."
+          },
+          {
+            "left": "issue:43257",
+            "right": "issue:43329",
+            "accept": false,
+            "reason": "Both mention Qwen3/MoE and multimodal code, but one is weight conversion during loading and the other is a video token-counting bug."
+          },
+          {
+            "left": "issue:43454",
+            "right": "issue:43577",
+            "accept": false,
+            "reason": "Both are model-loading related, but the failures are different: tied weights vs dtype propagation."
+          },
+          {
+            "left": "issue:43720",
+            "right": "issue:44610",
+            "accept": false,
+            "reason": "Packed-weight loading and image processor output shape are unrelated."
+          },
+          {
+            "left": "issue:43550",
+            "right": "issue:44912",
+            "accept": false,
+            "reason": "torch.compile/SDPA failure is unrelated to MXFP4 quantization loading behavior."
+          },
+          {
+            "left": "issue:44568",
+            "right": "issue:45488",
+            "accept": false,
+            "reason": "Both are tokenization regressions, but they affect different tokenizers and different mechanics (special tokens vs pre-tokenizer override)."
+          },
+          {
+            "left": "issue:43749",
+            "right": "issue:44291",
+            "accept": false,
+            "reason": "Both are loading-related, but FSDP CPU RAM loading and init_empty_weights have different failure modes and code paths."
+          },
+          {
+            "left": "issue:43531",
+            "right": "issue:45042",
+            "accept": false,
+            "reason": "Sliding-window behavior in Qwen3-MoE is unrelated to PIL/torchvision processor dependency issues."
+          },
+          {
+            "left": "issue:43262",
+            "right": "issue:44811",
+            "accept": false,
+            "reason": "Audio chat-template sampling-rate defaults and Whisper batch_decode token skipping are different processor behaviors."
+          },
+          {
+            "left": "issue:44568",
+            "right": "issue:45356",
+            "accept": false,
+            "reason": "Both are tokenizer regressions, but they are distinct bugs affecting different models and failure mechanisms."
+          },
+          {
+            "left": "issue:36010",
+            "right": "issue:36331",
+            "accept": false,
+            "reason": "ImportError from generation module and CustomTrainer signature breakage are unrelated."
+          },
+          {
+            "left": "issue:43643",
+            "right": "issue:45759",
+            "accept": false,
+            "reason": "Both touch config/model loading, but missing custom fields in AutoConfig is not the same as text_config unwrapping for multimodal causal LM creation."
+          },
+          {
+            "left": "issue:44849",
+            "right": "issue:45701",
+            "accept": false,
+            "reason": "Qwen3.5 hidden-state output regression is not a tokenization-version issue."
+          },
+          {
+            "left": "issue:34567",
+            "right": "issue:40990",
+            "accept": false,
+            "reason": "Trainer token counters and perplexity evaluation are unrelated."
+          },
+          {
+            "left": "issue:43329",
+            "right": "issue:43450",
+            "accept": false,
+            "reason": "Both are video-related, but one is token counting/undefined names and the other is batched processor output shape."
+          },
+          {
+            "left": "issue:43329",
+            "right": "issue:44479",
+            "accept": false,
+            "reason": "Both mention video regressions, but they appear to hit different layers: token counting vs model video-input regression."
+          },
+          {
+            "left": "issue:43606",
+            "right": "issue:45859",
+            "accept": false,
+            "reason": "Both are CPU-offload/device issues, but they affect different models and different failing operations."
+          },
+          {
+            "left": "issue:43257",
+            "right": "issue:43931",
+            "accept": false,
+            "reason": "Different loading failures: MOE weight conversion vs weight-shape mismatch."
+          },
+          {
+            "left": "issue:43697",
+            "right": "issue:45357",
+            "accept": false,
+            "reason": "Different regressions: v5 output drift for RTDetrV2 vs Qwen3.5 save_pretrained key corruption."
+          },
+          {
+            "left": "issue:45823",
+            "right": "issue:45859",
+            "accept": false,
+            "reason": "Both involve device/offload, but the underlying failures and model code paths are different."
+          },
+          {
+            "left": "issue:44291",
+            "right": "issue:45042",
+            "accept": false,
+            "reason": "init_empty_weights loading and PIL backend torchvision dependency are unrelated."
+          },
+          {
+            "left": "issue:41720",
+            "right": "issue:43540",
+            "accept": false,
+            "reason": "Both are Qwen3-family issues, but one is auto device mapping/cuda assert and the other is video-input validation in a different model."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 17,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:33357",
+        "issue:34567",
+        "issue:36010",
+        "issue:36331",
+        "issue:38175",
+        "issue:42175",
+        "issue:42915",
+        "issue:43257",
+        "issue:43317",
+        "issue:43452",
+        "issue:43493",
+        "issue:43540",
+        "issue:43577",
+        "issue:43582",
+        "issue:43643",
+        "issue:43646",
+        "issue:43696",
+        "issue:43742",
+        "issue:43749",
+        "issue:43756",
+        "issue:43824",
+        "issue:43906",
+        "issue:43927",
+        "issue:43994",
+        "issue:44062",
+        "issue:44291",
+        "issue:44479",
+        "issue:44488",
+        "issue:44560",
+        "issue:44737",
+        "issue:44756",
+        "issue:44792",
+        "issue:44821",
+        "issue:44898",
+        "issue:44912",
+        "issue:44964",
+        "issue:44987",
+        "issue:44991",
+        "issue:45072",
+        "issue:45081",
+        "issue:45137",
+        "issue:45375",
+        "issue:45381",
+        "issue:45405",
+        "issue:45538",
+        "issue:45588",
+        "issue:45823"
+      ],
+      "soft_pairs": [
+        "issue:43452|issue:43582",
+        "issue:42915|issue:43742",
+        "issue:43317|issue:45538",
+        "issue:33357|issue:34567",
+        "issue:43927|issue:44912",
+        "issue:44987|issue:45538",
+        "issue:42915|issue:44821",
+        "issue:44479|issue:45381",
+        "issue:42915|issue:45538",
+        "issue:43257|issue:43756",
+        "issue:43540|issue:44560",
+        "issue:43540|issue:44479",
+        "issue:43257|issue:43696",
+        "issue:43577|issue:45072",
+        "issue:45137|issue:45405",
+        "issue:38175|issue:44898",
+        "issue:44737|issue:45588",
+        "issue:44062|issue:44991",
+        "issue:43742|issue:44291",
+        "issue:43540|issue:45823",
+        "issue:43906|issue:45381",
+        "issue:43646|issue:43749",
+        "issue:44062|issue:45081",
+        "issue:43749|issue:44756",
+        "issue:34567|issue:36331",
+        "issue:43643|issue:45375",
+        "issue:36010|issue:42175",
+        "issue:43493|issue:43994",
+        "issue:42175|issue:43824",
+        "issue:43452|issue:44488",
+        "issue:44291|issue:44821",
+        "issue:44792|issue:44964"
+      ],
+      "prepared_review_unit_hash": "f2973b0975d256d41feea2602dfdce8172471a882fc5fe3413cce4f04bc08230",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 47,
+        "item_count": 47,
+        "soft_pair_count": 32,
+        "serialized_chars": 36765,
+        "estimated_input_tokens": 9192,
+        "estimated_eval_tokens": 18640
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "The packet is mostly heterogeneous. The only clear duplicate-worthy pairs are the Qwen video-input regression reports (44479/45381) and the SigLIP2 implementation/output reports (43493/43994). For the global issue anchor, 43824 is the stronger overall representative on the supplied activity totals, slightly ahead of 43749; I would avoid claiming a clearly reproducible regression from metadata alone.",
+        "confidence": 0.82,
+        "canonical_issue_reason": "Issue 44479 is the better duplicate anchor for the Qwen video-input pair because its title is the broader regression statement, while 45381 reads like a narrower symptom report for qwen2.5-vl video position IDs. That makes 44479 the more general anchor based on the packet text.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "Issue 43824 is the best overall issue anchor in this packet because it has the strongest combined engagement signal in the supplied metadata (discussion plus inbound references), and its import error spans the broader Qwen2.5-VL family rather than a narrower single-symptom report.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:43452",
+            "right": "issue:43582",
+            "accept": false,
+            "reason": "Tokenizer/model-loading failure vs Apple Silicon allocator warmup TypeError; different code paths and symptoms."
+          },
+          {
+            "left": "issue:42915",
+            "right": "issue:43742",
+            "accept": false,
+            "reason": "Qwen3Moe FP8 config failure vs MobileLLM key error; unrelated models and bugs."
+          },
+          {
+            "left": "issue:43317",
+            "right": "issue:45538",
+            "accept": false,
+            "reason": "Dequantized offload/device-map issue vs CLIPTokenizer max_length setting; not the same defect."
+          },
+          {
+            "left": "issue:33357",
+            "right": "issue:34567",
+            "accept": false,
+            "reason": "MacOS CLIP bus error vs TrainerState token counter not updating; different subsystems."
+          },
+          {
+            "left": "issue:43927",
+            "right": "issue:44912",
+            "accept": false,
+            "reason": "Config token-ID loss during save/load vs MXFP4 quantization fallback; distinct failures."
+          },
+          {
+            "left": "issue:44987",
+            "right": "issue:45538",
+            "accept": false,
+            "reason": "Model-loading regression vs tokenizer max_length behavior; unrelated."
+          },
+          {
+            "left": "issue:42915",
+            "right": "issue:44821",
+            "accept": false,
+            "reason": "FP8 Qwen3Moe issue vs AutoImageProcessor URL loading; different areas of the library."
+          },
+          {
+            "left": "issue:44479",
+            "right": "issue:45381",
+            "accept": true,
+            "reason": "Both describe the same Qwen video-input regression around incorrect video position IDs in the shared video code path."
+          },
+          {
+            "left": "issue:42915",
+            "right": "issue:45538",
+            "accept": false,
+            "reason": "Qwen3Moe FP8 config bug vs CLIPTokenizer length default; unrelated."
+          },
+          {
+            "left": "issue:43257",
+            "right": "issue:43756",
+            "accept": false,
+            "reason": "Qwen MOE weight conversion bug vs Smollm3 RoPE-layer mismatch; different model families and fixes."
+          },
+          {
+            "left": "issue:43540",
+            "right": "issue:44560",
+            "accept": false,
+            "reason": "Both are Qwen video issues, but the model families and concrete failures differ too much to treat as one bug."
+          },
+          {
+            "left": "issue:43540",
+            "right": "issue:44479",
+            "accept": false,
+            "reason": "Shared video theme, but different Qwen model families and failure modes; too broad to merge."
+          },
+          {
+            "left": "issue:43257",
+            "right": "issue:43696",
+            "accept": false,
+            "reason": "MOE conversion bug vs CUDA OOM during GPT-oss loading; not the same problem."
+          },
+          {
+            "left": "issue:43577",
+            "right": "issue:45072",
+            "accept": false,
+            "reason": "Different dtype mismatch reports in different models/code paths."
+          },
+          {
+            "left": "issue:45137",
+            "right": "issue:45405",
+            "accept": false,
+            "reason": "ZeRO3 deque error vs PEFT version pinning; unrelated."
+          },
+          {
+            "left": "issue:38175",
+            "right": "issue:44898",
+            "accept": false,
+            "reason": "SigLIP2 zero-probability issue vs Perceiver interpolation bug; different models and symptoms."
+          },
+          {
+            "left": "issue:44737",
+            "right": "issue:45588",
+            "accept": false,
+            "reason": "XLNet CPU tensor creation bug vs flash-attention sinkless-model crash; unrelated."
+          },
+          {
+            "left": "issue:44062",
+            "right": "issue:44991",
+            "accept": false,
+            "reason": "AddedToken kwarg duplication vs tokenizer load failure for a specific model; different causes."
+          },
+          {
+            "left": "issue:43742",
+            "right": "issue:44291",
+            "accept": false,
+            "reason": "MobileLLM key error vs init_empty_weights signature regression; different loading failures."
+          },
+          {
+            "left": "issue:43540",
+            "right": "issue:45823",
+            "accept": false,
+            "reason": "Qwen video-processing failure vs Gemma4 PLE device mismatch; unrelated."
+          },
+          {
+            "left": "issue:43906",
+            "right": "issue:45381",
+            "accept": false,
+            "reason": "Isolated reproduction of another issue vs a specific Qwen2.5-VL video-position bug; not the same defect."
+          },
+          {
+            "left": "issue:43646",
+            "right": "issue:43749",
+            "accept": false,
+            "reason": "Custom model initialization breakage vs FSDP CPU RAM-efficient loading; different code paths."
+          },
+          {
+            "left": "issue:44062",
+            "right": "issue:45081",
+            "accept": false,
+            "reason": "Different tokenizer bugs with different error mechanisms."
+          },
+          {
+            "left": "issue:43749",
+            "right": "issue:44756",
+            "accept": false,
+            "reason": "Both are loading/memory related, but one is FSDP CPU loading and the other is mmap on Strix Halo; different fixes."
+          },
+          {
+            "left": "issue:34567",
+            "right": "issue:36331",
+            "accept": false,
+            "reason": "TrainerState token counting bug vs custom compute_loss signature change; unrelated training APIs."
+          },
+          {
+            "left": "issue:43643",
+            "right": "issue:45375",
+            "accept": false,
+            "reason": "trust_remote_code field handling vs strict-dropped Qwen vision config field; different serialization paths."
+          },
+          {
+            "left": "issue:36010",
+            "right": "issue:42175",
+            "accept": false,
+            "reason": "GenerationMixin import regression vs TensorFlow backend packaging; unrelated."
+          },
+          {
+            "left": "issue:43493",
+            "right": "issue:43994",
+            "accept": true,
+            "reason": "Both are SigLIP2 reports pointing to the same underlying implementation mismatch and output-quality regression for the same model family."
+          },
+          {
+            "left": "issue:42175",
+            "right": "issue:43824",
+            "accept": false,
+            "reason": "Backend packaging issue vs Qwen import error; unrelated."
+          },
+          {
+            "left": "issue:43452",
+            "right": "issue:44488",
+            "accept": false,
+            "reason": "gguf_file loading breakage vs a specific model-load failure; different concrete causes."
+          },
+          {
+            "left": "issue:44291",
+            "right": "issue:44821",
+            "accept": false,
+            "reason": "init_empty_weights argument bug vs AutoImageProcessor URL loading; unrelated."
+          },
+          {
+            "left": "issue:44792",
+            "right": "issue:44964",
+            "accept": false,
+            "reason": "Janus image-generation test failure vs Phi-4 multimodal load failure; different models and symptoms."
+          }
+        ]
+      },
+      "evaluator_result": {
+        "accept": false,
+        "feedback": "The soft-edge verdicts are mostly conservative, but the top-level ranking overstates the packet. Issue 43749 is not clearly the most-discussed or most-referenced item in the set\u201443824 appears stronger by the provided activity counts\u2014and the phrase \"clear reproducible regression\" is not supported by the packet metadata. The canonical duplicate anchor choice for 44479 vs 45381 is plausible, but the representative-issue rationale needs to be more cautious and fully grounded in the supplied data."
+      },
+      "evaluator_used": true,
+      "retried": true,
+      "accepted_nontrivial_soft_edge": true,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 18,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:24643",
+        "issue:28282",
+        "issue:29127",
+        "issue:34689",
+        "issue:41762",
+        "issue:42175",
+        "issue:43295",
+        "issue:43452",
+        "issue:43526",
+        "issue:43531",
+        "issue:43577",
+        "issue:43582",
+        "issue:43643",
+        "issue:43653",
+        "issue:43697",
+        "issue:43720",
+        "issue:43742",
+        "issue:43756",
+        "issue:43819",
+        "issue:44466",
+        "issue:44484",
+        "issue:44514",
+        "issue:44561",
+        "issue:44568",
+        "issue:44589",
+        "issue:44625",
+        "issue:44704",
+        "issue:44811",
+        "issue:44821",
+        "issue:44843",
+        "issue:44877",
+        "issue:44898",
+        "issue:44936",
+        "issue:44938",
+        "issue:44977",
+        "issue:45005",
+        "issue:45042",
+        "issue:45081",
+        "issue:45216",
+        "issue:45290",
+        "issue:45325",
+        "issue:45356",
+        "issue:45381",
+        "issue:45405",
+        "issue:45479",
+        "issue:45600",
+        "issue:45823"
+      ],
+      "soft_pairs": [
+        "issue:43526|issue:45479",
+        "issue:28282|issue:34689",
+        "issue:45042|issue:45081",
+        "issue:43756|issue:45325",
+        "issue:43742|issue:44589",
+        "issue:45405|issue:45600",
+        "issue:43531|issue:44821",
+        "issue:43697|issue:45216",
+        "issue:44514|issue:45381",
+        "issue:28282|issue:29127",
+        "issue:44484|issue:44704",
+        "issue:43643|issue:44561",
+        "issue:43582|issue:44977",
+        "issue:44625|issue:44877",
+        "issue:43720|issue:44898",
+        "issue:42175|issue:44938",
+        "issue:43452|issue:43577",
+        "issue:41762|issue:44936",
+        "issue:44514|issue:45290",
+        "issue:43452|issue:44843",
+        "issue:44466|issue:45005",
+        "issue:43295|issue:45356",
+        "issue:43819|issue:44811",
+        "issue:43653|issue:44568",
+        "issue:44936|issue:45823",
+        "issue:24643|issue:29127"
+      ],
+      "prepared_review_unit_hash": "d0a4ac06784cd0031f239c440bcbf3d6296a765359c14a0c120631f1feb7df61",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 47,
+        "item_count": 47,
+        "soft_pair_count": 26,
+        "serialized_chars": 35370,
+        "estimated_input_tokens": 8843,
+        "estimated_eval_tokens": 17942
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "This cluster is not a duplicate group: the items span unrelated loading, tokenizer, processor, model-specific, and backend regression bugs. None of the soft-edge pairs look like the same underlying issue.",
+        "confidence": 0.98,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": null,
+        "best_issue_reason": null,
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:43526",
+            "right": "issue:45479",
+            "accept": false,
+            "reason": "Different bugs: BeitImageProcessorFast label reduction vs sequence-classification zero-loss behavior across many models."
+          },
+          {
+            "left": "issue:28282",
+            "right": "issue:34689",
+            "accept": false,
+            "reason": "Both are loading-related, but one is missing PyTorch detection and the other is a specific Llama 3.2 Vision regression."
+          },
+          {
+            "left": "issue:45042",
+            "right": "issue:45081",
+            "accept": false,
+            "reason": "Both touch image/tokenizer paths, but one is PIL/torchvision dependency handling and the other is a Mistral regex backend attribute crash."
+          },
+          {
+            "left": "issue:43756",
+            "right": "issue:45325",
+            "accept": false,
+            "reason": "Different model families and failures: Smollm3 RoPE-layer handling vs Qwen2.5-VL still-image temporal position IDs."
+          },
+          {
+            "left": "issue:43742",
+            "right": "issue:44589",
+            "accept": false,
+            "reason": "Different low-level failures: MobileLLM key error on load vs missing Float8 storage object error."
+          },
+          {
+            "left": "issue:45405",
+            "right": "issue:45600",
+            "accept": false,
+            "reason": "PEFT version bump issue is unrelated to auto_mappings referencing removed Sam3LiteText configs."
+          },
+          {
+            "left": "issue:43531",
+            "right": "issue:44821",
+            "accept": false,
+            "reason": "Qwen3-MoE sliding_window behavior is unrelated to AutoImageProcessor URL loading."
+          },
+          {
+            "left": "issue:43697",
+            "right": "issue:45216",
+            "accept": false,
+            "reason": "Different regressions: RTDetrV2 output mismatch in v5 vs Qwen3.5 checkpoint serialization correctness."
+          },
+          {
+            "left": "issue:44514",
+            "right": "issue:45381",
+            "accept": false,
+            "reason": "Different Qwen2.5-VL problems: chat-template batching crash vs video vision_position_ids calculation."
+          },
+          {
+            "left": "issue:28282",
+            "right": "issue:29127",
+            "accept": false,
+            "reason": "One is missing PyTorch in environment; the other is a LayoutLMv3 boxes validation/error-message issue."
+          },
+          {
+            "left": "issue:44484",
+            "right": "issue:44704",
+            "accept": false,
+            "reason": "save_pretrained shard-size behavior is unrelated to AutoProcessor kwargs not reaching cached_file."
+          },
+          {
+            "left": "issue:43643",
+            "right": "issue:44561",
+            "accept": false,
+            "reason": "Both involve v5/trust_remote_code compatibility, but one is missing config fields and the other is removal of is_torch_fx_available."
+          },
+          {
+            "left": "issue:43582",
+            "right": "issue:44977",
+            "accept": false,
+            "reason": "Apple Silicon warmup TypeError is unrelated to Qwen3.5 flash-attention generation issues."
+          },
+          {
+            "left": "issue:44625",
+            "right": "issue:44877",
+            "accept": false,
+            "reason": "Qwen3.5 num_labels propagation is a model config bug, not the strict granite_speech config-loading failure."
+          },
+          {
+            "left": "issue:43720",
+            "right": "issue:44898",
+            "accept": false,
+            "reason": "BitNet packed-weight unpacking during accelerate loading is unrelated to Perceiver interpolation/resolution failure."
+          },
+          {
+            "left": "issue:42175",
+            "right": "issue:44938",
+            "accept": false,
+            "reason": "Backend packaging mismatch for torch install is unrelated to Python 3.14 import/load failure."
+          },
+          {
+            "left": "issue:43452",
+            "right": "issue:43577",
+            "accept": false,
+            "reason": "gguf_file tokenizer/model loading breakage is unrelated to BLIP2 dtype propagation."
+          },
+          {
+            "left": "issue:41762",
+            "right": "issue:44936",
+            "accept": false,
+            "reason": "Gemma3 ZeRO-3 loading index error is unrelated to trainer.evaluate() failing after train()."
+          },
+          {
+            "left": "issue:44514",
+            "right": "issue:45290",
+            "accept": false,
+            "reason": "Both are chat-template issues, but one is batched padding behavior and the other is tool-call messages with empty content."
+          },
+          {
+            "left": "issue:43452",
+            "right": "issue:44843",
+            "accept": false,
+            "reason": "gguf_file loading bug is unrelated to offline mistral_regex patch calling model_info() unconditionally."
+          },
+          {
+            "left": "issue:44466",
+            "right": "issue:45005",
+            "accept": false,
+            "reason": "Both concern tied weights, but they are different serialization/save_pretrained regressions affecting different model classes."
+          },
+          {
+            "left": "issue:43295",
+            "right": "issue:45356",
+            "accept": false,
+            "reason": "Custom processor.tokenizer regression is unrelated to Kimi-K2.5 codec handling and mistral-regex warning."
+          },
+          {
+            "left": "issue:43819",
+            "right": "issue:44811",
+            "accept": false,
+            "reason": "DAC from_latents mismatch is unrelated to Whisper batch_decode ignoring skip_special_tokens."
+          },
+          {
+            "left": "issue:43653",
+            "right": "issue:44568",
+            "accept": false,
+            "reason": "BigBirdTokenizer special-token registration bug is unrelated to mdeberta-v3 add_special_tokens not adding BOS/EOS in v5."
+          },
+          {
+            "left": "issue:44936",
+            "right": "issue:45823",
+            "accept": false,
+            "reason": "trainer.evaluate() failure after train() is unrelated to Gemma4 device mismatch under device_map=\"auto\"."
+          },
+          {
+            "left": "issue:24643",
+            "right": "issue:29127",
+            "accept": false,
+            "reason": "DeepSpeed weight-dimension runtime error is unrelated to LayoutLMv3 insufficient-box-information validation."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 19,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:28282",
+        "issue:30064",
+        "issue:32090",
+        "issue:33666",
+        "issue:34567",
+        "issue:35141",
+        "issue:36010",
+        "issue:38175",
+        "issue:41720",
+        "issue:43122",
+        "issue:43295",
+        "issue:43329",
+        "issue:43452",
+        "issue:43504",
+        "issue:43525",
+        "issue:43531",
+        "issue:43550",
+        "issue:43572",
+        "issue:43577",
+        "issue:43653",
+        "issue:43688",
+        "issue:43697",
+        "issue:43720",
+        "issue:43742",
+        "issue:43749",
+        "issue:43756",
+        "issue:43906",
+        "issue:43927",
+        "issue:44361",
+        "issue:44442",
+        "issue:44451",
+        "issue:44464",
+        "issue:44568",
+        "issue:44610",
+        "issue:44743",
+        "issue:44821",
+        "issue:44843",
+        "issue:44849",
+        "issue:44913",
+        "issue:44977",
+        "issue:44991",
+        "issue:45325",
+        "issue:45356",
+        "issue:45357",
+        "issue:45440",
+        "issue:45588",
+        "issue:45812"
+      ],
+      "soft_pairs": [
+        "issue:30064|issue:33666",
+        "issue:38175|issue:43295",
+        "issue:44361|issue:44442",
+        "issue:43653|issue:44361",
+        "issue:43577|issue:43697",
+        "issue:44464|issue:44743",
+        "issue:44843|issue:44991",
+        "issue:41720|issue:43906",
+        "issue:44977|issue:45356",
+        "issue:43688|issue:45440",
+        "issue:43531|issue:43742",
+        "issue:43653|issue:43927",
+        "issue:44464|issue:44977",
+        "issue:43550|issue:43927",
+        "issue:43577|issue:45588",
+        "issue:43550|issue:44361",
+        "issue:32090|issue:34567",
+        "issue:43504|issue:43550",
+        "issue:43504|issue:43720",
+        "issue:43122|issue:43906",
+        "issue:28282|issue:35141",
+        "issue:32090|issue:33666",
+        "issue:43504|issue:44610",
+        "issue:43452|issue:44451",
+        "issue:44849|issue:45357",
+        "issue:44568|issue:45812",
+        "issue:43742|issue:44821",
+        "issue:41720|issue:44464",
+        "issue:33666|issue:36010",
+        "issue:43525|issue:43572",
+        "issue:43756|issue:44913",
+        "issue:43329|issue:45325",
+        "issue:43688|issue:43749"
+      ],
+      "prepared_review_unit_hash": "b909d741ef8fa533178917c30fb1cb5f5af960dae7c0036ae027dc369ac9c79c",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 47,
+        "item_count": 47,
+        "soft_pair_count": 33,
+        "serialized_chars": 37186,
+        "estimated_input_tokens": 9297,
+        "estimated_eval_tokens": 18850
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "The candidate pairs are mostly superficial matches by subsystem only; none appear to describe the same underlying bug or change. This looks like a mixed set of unrelated issues rather than one duplicate cluster.",
+        "confidence": 0.9,
+        "canonical_issue_reason": "There is no single canonical duplicate here because the issues span many unrelated areas. If forced to pick the broadest representative, issue 45812 is the most general tokenizer regression, but it still does not represent the rest of the set.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "Issue 45812 is the strongest global representative only in the narrow sense that it is a broad, open tokenizer regression affecting multiple Granite models. Even so, it is not a true canonical for this heterogeneous group.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:30064",
+            "right": "issue:33666",
+            "accept": false,
+            "reason": "Image processor segmentation-map failure and Qwen2-VL multi-GPU training are different components and different failure modes."
+          },
+          {
+            "left": "issue:38175",
+            "right": "issue:43295",
+            "accept": false,
+            "reason": "Zero-probabilities in SigLIP2 and a processor/tokenizer regression in v4.57.5 are unrelated bugs."
+          },
+          {
+            "left": "issue:44361",
+            "right": "issue:44442",
+            "accept": false,
+            "reason": "MLukeTokenizer attribute error and AutoTokenizer failing on FastSpeech2ConformerTokenizer are distinct tokenizer-loading issues."
+          },
+          {
+            "left": "issue:43653",
+            "right": "issue:44361",
+            "accept": false,
+            "reason": "BigBird mask-token registration and MLukeTokenizer task failures do not describe the same bug."
+          },
+          {
+            "left": "issue:43577",
+            "right": "issue:43697",
+            "accept": false,
+            "reason": "BLIP2 dtype handling and RTDetrV2 output drift are separate model behavior problems."
+          },
+          {
+            "left": "issue:44464",
+            "right": "issue:44743",
+            "accept": false,
+            "reason": "Chunked generation with compiled forward and Qwen3.5 recurrent-state reset are different generation bugs."
+          },
+          {
+            "left": "issue:44843",
+            "right": "issue:44991",
+            "accept": false,
+            "reason": "HF_HUB_OFFLINE/model_info patching and tokenizer loading for est-roberta are different tokenizer path regressions."
+          },
+          {
+            "left": "issue:41720",
+            "right": "issue:43906",
+            "accept": false,
+            "reason": "Qwen3 auto device mapping CUDA assert and an isolated reproduction of another issue are not the same concrete failure."
+          },
+          {
+            "left": "issue:44977",
+            "right": "issue:45356",
+            "accept": false,
+            "reason": "Flash-attention generation failure and Kimi-K2.5 tokenizer codec/regex regression are unrelated."
+          },
+          {
+            "left": "issue:43688",
+            "right": "issue:45440",
+            "accept": false,
+            "reason": "Aux-loss normalization and DeepseekV3MoE implementation divergence are different algorithmic issues."
+          },
+          {
+            "left": "issue:43531",
+            "right": "issue:43742",
+            "accept": false,
+            "reason": "Qwen3-MoE sliding_window behavior and MobileLLM loading key errors are unrelated."
+          },
+          {
+            "left": "issue:43653",
+            "right": "issue:43927",
+            "accept": false,
+            "reason": "BigBird special-token registration and DiaConfig custom token-ID loss on save/load are different tokenizer/config bugs."
+          },
+          {
+            "left": "issue:44464",
+            "right": "issue:44977",
+            "accept": false,
+            "reason": "Compiled chunked generation inconsistency is not the same as Qwen3.5 flash-attention generation failure."
+          },
+          {
+            "left": "issue:43550",
+            "right": "issue:43927",
+            "accept": false,
+            "reason": "Bamba torch.compile/SDPA failure and DiaConfig token-ID persistence are different failure modes."
+          },
+          {
+            "left": "issue:43577",
+            "right": "issue:45588",
+            "accept": false,
+            "reason": "BLIP2 dtype loading and flash_attention.py crashing on s_aux=None are unrelated."
+          },
+          {
+            "left": "issue:43550",
+            "right": "issue:44361",
+            "accept": false,
+            "reason": "Bamba compile/SDPA and MLukeTokenizer attribute errors are unrelated areas and symptoms."
+          },
+          {
+            "left": "issue:32090",
+            "right": "issue:34567",
+            "accept": false,
+            "reason": "Trainer broadcast NoneType error and num_input_tokens_seen not updating are separate trainer-state bugs."
+          },
+          {
+            "left": "issue:43504",
+            "right": "issue:43550",
+            "accept": false,
+            "reason": "Beit legacy-field loading and Bamba torch.compile/SDPA are different code paths and bugs."
+          },
+          {
+            "left": "issue:43504",
+            "right": "issue:43720",
+            "accept": false,
+            "reason": "Beit legacy-field load failure and BitNet packed-weight unpacking are unrelated loading issues."
+          },
+          {
+            "left": "issue:43122",
+            "right": "issue:43906",
+            "accept": false,
+            "reason": "Tokenizer behavior change across versions and an isolated reproduction of another issue are not enough to identify the same bug."
+          },
+          {
+            "left": "issue:28282",
+            "right": "issue:35141",
+            "accept": false,
+            "reason": "Missing PyTorch ImportError and output embedding reinitialization after resize are unrelated."
+          },
+          {
+            "left": "issue:32090",
+            "right": "issue:33666",
+            "accept": false,
+            "reason": "NoneType broadcast in Trainer and Qwen2-VL multi-GPU training are different training problems."
+          },
+          {
+            "left": "issue:43504",
+            "right": "issue:44610",
+            "accept": false,
+            "reason": "Beit pretrained-load legacy field issue and OmDet-Turbo input-size mismatch are unrelated vision bugs."
+          },
+          {
+            "left": "issue:43452",
+            "right": "issue:44451",
+            "accept": false,
+            "reason": "gguf_file handling for AutoTokenizer/AutoModelForCausalLM and inability to load ScandiBERT are distinct load failures."
+          },
+          {
+            "left": "issue:44849",
+            "right": "issue:45357",
+            "accept": false,
+            "reason": "Qwen3.5 hidden-state bug and incorrect visual-encoder keys on save_pretrained are different regressions."
+          },
+          {
+            "left": "issue:44568",
+            "right": "issue:45812",
+            "accept": false,
+            "reason": "mdeberta special-token insertion and Granite token-ID regression are both tokenizer issues but not the same model-specific bug."
+          },
+          {
+            "left": "issue:43742",
+            "right": "issue:44821",
+            "accept": false,
+            "reason": "MobileLLM key error on load and AutoImageProcessor-from-URL failure are unrelated loading paths."
+          },
+          {
+            "left": "issue:41720",
+            "right": "issue:44464",
+            "accept": false,
+            "reason": "Qwen3 device-mapping CUDA assert and compiled chunked generation inconsistency are different runtime failures."
+          },
+          {
+            "left": "issue:33666",
+            "right": "issue:36010",
+            "accept": false,
+            "reason": "Qwen2-VL multi-GPU training and GenerationMixin import error are unrelated."
+          },
+          {
+            "left": "issue:43525",
+            "right": "issue:43572",
+            "accept": false,
+            "reason": "Llama4Config missing pad_token_id and StableLmConfig missing pad_token_idx are similar in theme but different model/config bugs."
+          },
+          {
+            "left": "issue:43756",
+            "right": "issue:44913",
+            "accept": false,
+            "reason": "Smollm3 RoPE-layer dropping and GPTNeoX rotary_pct reload defaulting are distinct config issues."
+          },
+          {
+            "left": "issue:43329",
+            "right": "issue:45325",
+            "accept": false,
+            "reason": "Undefined video-branch helpers in multimodal token counting and Qwen2.5-VL still-image position scaling are different RoPE/multimodal bugs."
+          },
+          {
+            "left": "issue:43688",
+            "right": "issue:43749",
+            "accept": false,
+            "reason": "Aux-loss normalization and broken FSDP CPU RAM efficient loading are unrelated."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 20,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:28282",
+        "issue:32090",
+        "issue:33357",
+        "issue:33666",
+        "issue:36331",
+        "issue:39692",
+        "issue:43257",
+        "issue:43262",
+        "issue:43278",
+        "issue:43295",
+        "issue:43299",
+        "issue:43334",
+        "issue:43493",
+        "issue:43504",
+        "issue:43540",
+        "issue:43572",
+        "issue:43577",
+        "issue:43606",
+        "issue:43646",
+        "issue:43688",
+        "issue:43720",
+        "issue:43756",
+        "issue:43824",
+        "issue:43825",
+        "issue:43844",
+        "issue:43927",
+        "issue:44291",
+        "issue:44361",
+        "issue:44423",
+        "issue:44466",
+        "issue:44479",
+        "issue:44488",
+        "issue:44554",
+        "issue:44589",
+        "issue:44625",
+        "issue:44734",
+        "issue:44843",
+        "issue:44964",
+        "issue:44977",
+        "issue:45083",
+        "issue:45200",
+        "issue:45229",
+        "issue:45405",
+        "issue:45520",
+        "issue:45701",
+        "issue:45704",
+        "issue:45759",
+        "issue:45823"
+      ],
+      "soft_pairs": [
+        "issue:43493|issue:43825",
+        "issue:43262|issue:43295",
+        "issue:43606|issue:43927",
+        "issue:43720|issue:43927",
+        "issue:28282|issue:32090",
+        "issue:28282|issue:33357",
+        "issue:43577|issue:44554",
+        "issue:33666|issue:36331",
+        "issue:43720|issue:44361",
+        "issue:39692|issue:43493",
+        "issue:43688|issue:45704",
+        "issue:43334|issue:43572",
+        "issue:43540|issue:44977",
+        "issue:44964|issue:45405",
+        "issue:44479|issue:44625",
+        "issue:43493|issue:43844",
+        "issue:43756|issue:45083",
+        "issue:45200|issue:45229",
+        "issue:43334|issue:45759",
+        "issue:44488|issue:44843",
+        "issue:43257|issue:43299",
+        "issue:43278|issue:43844",
+        "issue:43824|issue:44589",
+        "issue:43646|issue:44466",
+        "issue:45229|issue:45823",
+        "issue:44589|issue:45520",
+        "issue:43688|issue:45701",
+        "issue:39692|issue:43295",
+        "issue:43572|issue:44291",
+        "issue:33357|issue:33666",
+        "issue:43504|issue:43606",
+        "issue:44423|issue:44734",
+        "issue:43927|issue:44361"
+      ],
+      "prepared_review_unit_hash": "8a826ea6616c851b6136ceca47fd6d31e00541cdbc8940dc62a9b995a6abac2c",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 33,
+        "serialized_chars": 37972,
+        "estimated_input_tokens": 9493,
+        "estimated_eval_tokens": 19242
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "The cluster is highly heterogeneous: the soft-linked pairs mostly share only broad Transformer subsystems or model families, not the same concrete bug. No PRs are present.",
+        "confidence": 0.96,
+        "canonical_issue_reason": "issue:43824 is the most prominent standalone issue in the set by discussion and inbound references, with a concrete reproducible loading failure; the rest of the cluster is too diverse to serve as a true duplicate canonical.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "issue:43824 is the best overall issue representative if one must be chosen, because it has the strongest visibility and a clearly scoped bug report; however, it is not a duplicate of most other items here.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:43493",
+            "right": "issue:43825",
+            "accept": false,
+            "reason": "Different bugs: SigLIP2 implementation mismatch vs a pipeline error-message regression about translation support."
+          },
+          {
+            "left": "issue:43262",
+            "right": "issue:43295",
+            "accept": false,
+            "reason": "Different code paths: audio processor sampling-rate default vs processor.tokenizer/image passthrough regression."
+          },
+          {
+            "left": "issue:43606",
+            "right": "issue:43927",
+            "accept": false,
+            "reason": "Unrelated failures: CPU offload device mismatch vs DiaConfig custom token IDs lost on save/load."
+          },
+          {
+            "left": "issue:43720",
+            "right": "issue:43927",
+            "accept": false,
+            "reason": "Different issues: BitNet packed-weight handling during accelerate loading vs DiaConfig token ID persistence."
+          },
+          {
+            "left": "issue:28282",
+            "right": "issue:32090",
+            "accept": false,
+            "reason": "Import-time PyTorch missing error is unrelated to Trainer NoneType broadcast failure."
+          },
+          {
+            "left": "issue:28282",
+            "right": "issue:33357",
+            "accept": false,
+            "reason": "PyTorch availability error and MacOS CLIP bus error are distinct bugs."
+          },
+          {
+            "left": "issue:43577",
+            "right": "issue:44554",
+            "accept": false,
+            "reason": "BLIP2 dtype persistence and an MPS attention correctness bug are different concrete problems."
+          },
+          {
+            "left": "issue:33666",
+            "right": "issue:36331",
+            "accept": false,
+            "reason": "Multi-GPU Qwen2-VL training request is unrelated to the custom compute_loss signature error."
+          },
+          {
+            "left": "issue:43720",
+            "right": "issue:44361",
+            "accept": false,
+            "reason": "BitNet packed-weight loading failure and MLukeTokenizer task AttributeError are unrelated."
+          },
+          {
+            "left": "issue:39692",
+            "right": "issue:43493",
+            "accept": false,
+            "reason": "Docs/example mistakes are not the same as the SigLIP2 HF-vs-JAX implementation discrepancy."
+          },
+          {
+            "left": "issue:43688",
+            "right": "issue:45704",
+            "accept": false,
+            "reason": "Auxiliary-loss normalization and T5 apex RMSNorm memory leak are different model bugs."
+          },
+          {
+            "left": "issue:43334",
+            "right": "issue:43572",
+            "accept": false,
+            "reason": "Both involve config fields, but they are different models and different missing fields, not the same bug."
+          },
+          {
+            "left": "issue:43540",
+            "right": "issue:44977",
+            "accept": false,
+            "reason": "Qwen3OmniMoe video-input processing and Qwen3.5 flash-attention generation are separate failures."
+          },
+          {
+            "left": "issue:44964",
+            "right": "issue:45405",
+            "accept": false,
+            "reason": "Model load failure and a bumped unreleased dependency requirement are unrelated."
+          },
+          {
+            "left": "issue:44479",
+            "right": "issue:44625",
+            "accept": false,
+            "reason": "Video-input regression across Qwen variants is unrelated to num_labels propagation in Qwen3.5."
+          },
+          {
+            "left": "issue:43493",
+            "right": "issue:43844",
+            "accept": false,
+            "reason": "SigLIP2 implementation drift and DeepSpeed gradient growth are different bugs."
+          },
+          {
+            "left": "issue:43756",
+            "right": "issue:45083",
+            "accept": false,
+            "reason": "Smollm3 RoPE-layer drop and qwen3_omni_moe feature-length helper behavior are unrelated."
+          },
+          {
+            "left": "issue:45200",
+            "right": "issue:45229",
+            "accept": false,
+            "reason": "Default mm_token_type_ids behavior and Gemma4 multi-GPU OOM are not the same underlying issue."
+          },
+          {
+            "left": "issue:43334",
+            "right": "issue:45759",
+            "accept": false,
+            "reason": "Both involve Qwen multimodal configs, but one is a missing pad_token_id error and the other is from_config not unwrapping text_config."
+          },
+          {
+            "left": "issue:44488",
+            "right": "issue:44843",
+            "accept": false,
+            "reason": "Model load failure for a specific checkpoint is unrelated to HF_HUB_OFFLINE behavior in tokenizer patching."
+          },
+          {
+            "left": "issue:43257",
+            "right": "issue:43299",
+            "accept": false,
+            "reason": "Different Qwen MOE loading regressions with different failure modes and code paths."
+          },
+          {
+            "left": "issue:43278",
+            "right": "issue:43844",
+            "accept": false,
+            "reason": "Embedding dtype drift in train/eval is unrelated to the ZeRO-3 gradient increase report."
+          },
+          {
+            "left": "issue:43824",
+            "right": "issue:44589",
+            "accept": false,
+            "reason": "Missing Qwen2_5_VL import and Float8 storage lookup failure are unrelated load-time errors."
+          },
+          {
+            "left": "issue:43646",
+            "right": "issue:44466",
+            "accept": false,
+            "reason": "Custom model initialization breakage and lm_head.weight serialization inconsistency are separate issues."
+          },
+          {
+            "left": "issue:45229",
+            "right": "issue:45823",
+            "accept": false,
+            "reason": "Gemma4 inference OOM and PLE device mismatch are different forward-pass failures."
+          },
+          {
+            "left": "issue:44589",
+            "right": "issue:45520",
+            "accept": false,
+            "reason": "Float8 storage lookup error is unrelated to the flash_attn KeyError on Python 3.13."
+          },
+          {
+            "left": "issue:43688",
+            "right": "issue:45701",
+            "accept": false,
+            "reason": "Auxiliary-loss normalization and tokenization changes across versions are not the same bug."
+          },
+          {
+            "left": "issue:39692",
+            "right": "issue:43295",
+            "accept": false,
+            "reason": "Documentation example errors are unrelated to the processor.tokenizer regression in v4.57.5."
+          },
+          {
+            "left": "issue:43572",
+            "right": "issue:44291",
+            "accept": false,
+            "reason": "Missing pad_token_idx in StableLmConfig is a different problem from init_empty_weights passing _is_hf_initialized."
+          },
+          {
+            "left": "issue:33357",
+            "right": "issue:33666",
+            "accept": false,
+            "reason": "MacOS bus error with a CLIP model and Qwen2-VL multi-GPU training are unrelated."
+          },
+          {
+            "left": "issue:43504",
+            "right": "issue:43606",
+            "accept": false,
+            "reason": "Legacy-field load failure in BEiT is unrelated to the Bark device-mismatch CPU offload bug."
+          },
+          {
+            "left": "issue:44423",
+            "right": "issue:44734",
+            "accept": false,
+            "reason": "Continuous-batching multimodal crash and KV-cache continuation indexing bug are distinct serve-path issues."
+          },
+          {
+            "left": "issue:43927",
+            "right": "issue:44361",
+            "accept": false,
+            "reason": "DiaConfig save/load token-ID loss and MLukeTokenizer task AttributeError do not describe the same failure."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 21,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:15354",
+        "issue:28218",
+        "issue:29942",
+        "issue:30333",
+        "issue:33357",
+        "issue:34567",
+        "issue:38175",
+        "issue:43232",
+        "issue:43278",
+        "issue:43295",
+        "issue:43504",
+        "issue:43540",
+        "issue:43575",
+        "issue:43606",
+        "issue:43653",
+        "issue:43688",
+        "issue:43701",
+        "issue:43844",
+        "issue:43873",
+        "issue:43994",
+        "issue:44112",
+        "issue:44315",
+        "issue:44464",
+        "issue:44466",
+        "issue:44493",
+        "issue:44521",
+        "issue:44554",
+        "issue:44568",
+        "issue:44610",
+        "issue:44625",
+        "issue:44792",
+        "issue:44849",
+        "issue:44964",
+        "issue:45072",
+        "issue:45084",
+        "issue:45127",
+        "issue:45198",
+        "issue:45200",
+        "issue:45290",
+        "issue:45325",
+        "issue:45356",
+        "issue:45405",
+        "issue:45446",
+        "issue:45520",
+        "issue:45584",
+        "issue:45701",
+        "issue:45759",
+        "issue:45823"
+      ],
+      "soft_pairs": [
+        "issue:43504|issue:43653",
+        "issue:43575|issue:45405",
+        "issue:44625|issue:45759",
+        "issue:43606|issue:44112",
+        "issue:44792|issue:45405",
+        "issue:43994|issue:44493",
+        "issue:44521|issue:45290",
+        "issue:44464|issue:45584",
+        "issue:29942|issue:30333",
+        "issue:43844|issue:44964",
+        "issue:28218|issue:34567",
+        "issue:45084|issue:45520",
+        "issue:44554|issue:45446",
+        "issue:43688|issue:43873",
+        "issue:44568|issue:45701",
+        "issue:15354|issue:38175",
+        "issue:43232|issue:44849",
+        "issue:43540|issue:45200",
+        "issue:45200|issue:45823",
+        "issue:45072|issue:45198",
+        "issue:44315|issue:44849",
+        "issue:43278|issue:43701",
+        "issue:28218|issue:33357",
+        "issue:44554|issue:45356",
+        "issue:45200|issue:45325",
+        "issue:44466|issue:45127",
+        "issue:43295|issue:44610"
+      ],
+      "prepared_review_unit_hash": "c391cf41884d7611a219891571e059c421c53f2ecc555dd738a1e1fd4216f75d",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 27,
+        "serialized_chars": 35870,
+        "estimated_input_tokens": 8968,
+        "estimated_eval_tokens": 18192
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "Mostly unrelated issues spanning tokenization, generation, multimodal preprocessing, training, and CI. Only one soft edge looks like a plausible duplicate pair: the Qwen composite-config propagation/from_config handling issues.",
+        "confidence": 0.79,
+        "canonical_issue_reason": "issue:45759 is the best canonical issue: it states the broader root cause in composite Qwen multimodal config handling, while issue:44625 is a narrower symptom of the same config-propagation path.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "issue:45759 is the strongest representative because it describes the underlying API/config plumbing problem most clearly and broadly.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:43504",
+            "right": "issue:43653",
+            "accept": false,
+            "reason": "Different subsystems and failures: BEiT legacy config loading vs BigBird tokenizer special-token registration."
+          },
+          {
+            "left": "issue:43575",
+            "right": "issue:45405",
+            "accept": false,
+            "reason": "One is tensor-parallel OOM on a model load; the other is a dependency version bump problem. No shared bug."
+          },
+          {
+            "left": "issue:44625",
+            "right": "issue:45759",
+            "accept": true,
+            "reason": "Both point to the same composite Qwen config plumbing in model/config construction; one is a specific propagation symptom, the other the broader from_config unwrap bug."
+          },
+          {
+            "left": "issue:43606",
+            "right": "issue:44112",
+            "accept": false,
+            "reason": "Both mention device mismatches, but they affect different models and test paths; not the same underlying fix."
+          },
+          {
+            "left": "issue:44792",
+            "right": "issue:45405",
+            "accept": false,
+            "reason": "Janus image-generation test failure is unrelated to the PEFT minimum-version packaging issue."
+          },
+          {
+            "left": "issue:43994",
+            "right": "issue:44493",
+            "accept": false,
+            "reason": "Different symptoms and code paths: SigLIP2 wrong outputs vs position-id key handling."
+          },
+          {
+            "left": "issue:44521",
+            "right": "issue:45290",
+            "accept": false,
+            "reason": "Both involve apply_chat_template, but one is multimodal assistant-mask generation and the other is a crash on tool-call/no-content messages; not clearly the same bug."
+          },
+          {
+            "left": "issue:44464",
+            "right": "issue:45584",
+            "accept": false,
+            "reason": "Compiled-forward chunked generation inconsistency is unrelated to Whisper empty-transcription handling after token alignment."
+          },
+          {
+            "left": "issue:29942",
+            "right": "issue:30333",
+            "accept": false,
+            "reason": "Flash Attention CI failures and MLflow reporting on failed jobs are unrelated."
+          },
+          {
+            "left": "issue:43844",
+            "right": "issue:44964",
+            "accept": false,
+            "reason": "ZeRO-3 gradient growth and Phi-4 multimodal load failure are different problems."
+          },
+          {
+            "left": "issue:28218",
+            "right": "issue:34567",
+            "accept": false,
+            "reason": "Tokenizer added-space behavior and TrainerState token counter updates are unrelated."
+          },
+          {
+            "left": "issue:45084",
+            "right": "issue:45520",
+            "accept": false,
+            "reason": "Template compilation error and Python 3.13 flash_attn import KeyError are different issues."
+          },
+          {
+            "left": "issue:44554",
+            "right": "issue:45446",
+            "accept": false,
+            "reason": "MPS attention correctness and a PyTorch version check bug in flex_attention are not the same code path."
+          },
+          {
+            "left": "issue:43688",
+            "right": "issue:43873",
+            "accept": false,
+            "reason": "Aux-loss normalization and quantization/offload behavior are separate model/runtime issues."
+          },
+          {
+            "left": "issue:44568",
+            "right": "issue:45701",
+            "accept": false,
+            "reason": "Tokenizer special-token regression and generic tokenization changes across versions are too broad to treat as the same bug."
+          },
+          {
+            "left": "issue:15354",
+            "right": "issue:38175",
+            "accept": false,
+            "reason": "TorchScript export support and zero-probability outputs from SigLIP2 are unrelated."
+          },
+          {
+            "left": "issue:43232",
+            "right": "issue:44849",
+            "accept": false,
+            "reason": "Generation kwargs update after sync_gpus and Qwen3.5 hidden-state output behavior are different bugs."
+          },
+          {
+            "left": "issue:43540",
+            "right": "issue:45200",
+            "accept": false,
+            "reason": "Video-input processing in Qwen3OmniMoe and Gemma4 text-only token-type defaults are unrelated."
+          },
+          {
+            "left": "issue:45200",
+            "right": "issue:45823",
+            "accept": false,
+            "reason": "Same model family, but one is missing default mm_token_type_ids and the other is a device-map forward mismatch; not the same fix."
+          },
+          {
+            "left": "issue:45072",
+            "right": "issue:45198",
+            "accept": false,
+            "reason": "Dtype mismatch inference failures and Wav2Vec2 save/tokenization failures are unrelated."
+          },
+          {
+            "left": "issue:44315",
+            "right": "issue:44849",
+            "accept": false,
+            "reason": "Liger Kernel application with model_init is unrelated to Qwen3.5 hidden-state output behavior."
+          },
+          {
+            "left": "issue:43278",
+            "right": "issue:43701",
+            "accept": false,
+            "reason": "Training/eval dtype drift and resume_from_checkpoint key mismatch are separate training-state problems."
+          },
+          {
+            "left": "issue:28218",
+            "right": "issue:33357",
+            "accept": false,
+            "reason": "Token spacing after added tokens and a MacOS bus error in CLIP loading are unrelated."
+          },
+          {
+            "left": "issue:44554",
+            "right": "issue:45356",
+            "accept": false,
+            "reason": "MPS attention correctness and Kimi tokenizer codec/regression handling are different concerns."
+          },
+          {
+            "left": "issue:45200",
+            "right": "issue:45325",
+            "accept": false,
+            "reason": "Gemma4 token-type defaults and Qwen2.5-VL rope position scaling are different multimodal bugs."
+          },
+          {
+            "left": "issue:44466",
+            "right": "issue:45127",
+            "accept": false,
+            "reason": "Serialization of tied weights and LoRA merge collapse with extended vocabulary are not the same bug."
+          },
+          {
+            "left": "issue:43295",
+            "right": "issue:44610",
+            "accept": false,
+            "reason": "Processor/tokenizer regression and OmDet-Turbo image-size mismatch are unrelated."
+          }
+        ]
+      },
+      "evaluator_result": {
+        "accept": true,
+        "feedback": "The summary is grounded in the packet and stays conservative. The only accepted soft edge, issue:44625 vs issue:45759, is a plausible duplicate based on both titles pointing to Qwen composite config propagation/from_config handling. The rest of the rejects are cautious and do not overstate shared causality."
+      },
+      "evaluator_used": true,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": true,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 22,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:37428",
+        "issue:38175",
+        "issue:42915",
+        "issue:43262",
+        "issue:43278",
+        "issue:43295",
+        "issue:43334",
+        "issue:43425",
+        "issue:43450",
+        "issue:43493",
+        "issue:43525",
+        "issue:43572",
+        "issue:43575",
+        "issue:43653",
+        "issue:43701",
+        "issue:43819",
+        "issue:43824",
+        "issue:43825",
+        "issue:43827",
+        "issue:43844",
+        "issue:43856",
+        "issue:43873",
+        "issue:43901",
+        "issue:43957",
+        "issue:43975",
+        "issue:44112",
+        "issue:44206",
+        "issue:44360",
+        "issue:44442",
+        "issue:44464",
+        "issue:44514",
+        "issue:44521",
+        "issue:44561",
+        "issue:44610",
+        "issue:44792",
+        "issue:44857",
+        "issue:44898",
+        "issue:44936",
+        "issue:44987",
+        "issue:45083",
+        "issue:45127",
+        "issue:45198",
+        "issue:45245",
+        "issue:45412",
+        "issue:45563",
+        "issue:45584",
+        "issue:45593"
+      ],
+      "soft_pairs": [
+        "issue:44898|issue:45412",
+        "issue:43493|issue:43701",
+        "issue:37428|issue:43957",
+        "issue:43572|issue:44987",
+        "issue:43334|issue:43525",
+        "issue:43572|issue:43957",
+        "issue:43873|issue:44464",
+        "issue:43295|issue:43450",
+        "issue:43844|issue:44936",
+        "issue:43824|issue:43975",
+        "issue:43295|issue:43873",
+        "issue:38175|issue:45412",
+        "issue:37428|issue:43824",
+        "issue:42915|issue:43824",
+        "issue:44514|issue:44521",
+        "issue:38175|issue:43493",
+        "issue:44442|issue:45198",
+        "issue:43278|issue:43856",
+        "issue:43827|issue:43901",
+        "issue:43653|issue:44112",
+        "issue:43450|issue:43493",
+        "issue:43450|issue:43575",
+        "issue:44792|issue:45563",
+        "issue:45127|issue:45245",
+        "issue:44360|issue:44561",
+        "issue:43819|issue:45083",
+        "issue:44206|issue:45584",
+        "issue:43819|issue:44112",
+        "issue:43262|issue:43450",
+        "issue:44610|issue:44857",
+        "issue:37428|issue:43425",
+        "issue:44857|issue:45593",
+        "issue:43278|issue:43825"
+      ],
+      "prepared_review_unit_hash": "79beb0f0305d5fdef32a6795d8f8bfc594314295aad690874d07040af09fdb25",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 47,
+        "item_count": 47,
+        "soft_pair_count": 33,
+        "serialized_chars": 36920,
+        "estimated_input_tokens": 9230,
+        "estimated_eval_tokens": 18716
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "This cluster is mostly a grab-bag of unrelated issues across model loading, processors, training, docs, and CI. The soft pairs share broad keywords in some cases, but they do not look like the same underlying bug or change.",
+        "confidence": 0.96,
+        "canonical_issue_reason": "Issue 43824 is the strongest standalone anchor: it has a clear, reproducible import traceback, multiple references, and a focused regression signature.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "Issue 43824 is the best representative issue because it is concrete, well-discussed, and easy to triage or search by its exact failing import path.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:44898",
+            "right": "issue:45412",
+            "accept": false,
+            "reason": "Different subsystems and failures: Perceiver resolution handling vs RT-DETR memory release."
+          },
+          {
+            "left": "issue:43493",
+            "right": "issue:43701",
+            "accept": false,
+            "reason": "Different bugs: SigLIP2 implementation discrepancy vs checkpoint key mismatch."
+          },
+          {
+            "left": "issue:37428",
+            "right": "issue:43957",
+            "accept": false,
+            "reason": "Both are loading-related, but they describe different root causes and model paths."
+          },
+          {
+            "left": "issue:43572",
+            "right": "issue:44987",
+            "accept": false,
+            "reason": "Unrelated: missing pad_token_idx in one config vs loading failure for a specific model repo."
+          },
+          {
+            "left": "issue:43334",
+            "right": "issue:43525",
+            "accept": false,
+            "reason": "Same symptom class, but different model configs and no evidence of one shared code defect."
+          },
+          {
+            "left": "issue:43572",
+            "right": "issue:43957",
+            "accept": false,
+            "reason": "Different issues: tokenizer config field regression vs meta-device loading problems."
+          },
+          {
+            "left": "issue:43873",
+            "right": "issue:44464",
+            "accept": false,
+            "reason": "Different areas: quantization/offloading vs compiled-forward chunked generation consistency."
+          },
+          {
+            "left": "issue:43295",
+            "right": "issue:43450",
+            "accept": false,
+            "reason": "Different modalities and failures: processor/tokenizer regression vs batched video-shape bug."
+          },
+          {
+            "left": "issue:43844",
+            "right": "issue:44936",
+            "accept": false,
+            "reason": "Training instability with ZeRO-3 is unrelated to trainer.evaluate() breaking after train()."
+          },
+          {
+            "left": "issue:43824",
+            "right": "issue:43975",
+            "accept": false,
+            "reason": "Different models and failure modes: import error vs detokenization regression."
+          },
+          {
+            "left": "issue:43295",
+            "right": "issue:43873",
+            "accept": false,
+            "reason": "Processor API regression and quantization offloading are unrelated."
+          },
+          {
+            "left": "issue:38175",
+            "right": "issue:45412",
+            "accept": false,
+            "reason": "Different components: SigLIP2 zero probabilities vs RT-DETR memory leak."
+          },
+          {
+            "left": "issue:37428",
+            "right": "issue:43824",
+            "accept": false,
+            "reason": "Both are import errors, but they involve different symbols, modules, and root causes."
+          },
+          {
+            "left": "issue:42915",
+            "right": "issue:43824",
+            "accept": false,
+            "reason": "Different model families and different failure signatures."
+          },
+          {
+            "left": "issue:44514",
+            "right": "issue:44521",
+            "accept": false,
+            "reason": "Same general processor area, but one crashes on batching and the other returns wrong masks; not the same bug."
+          },
+          {
+            "left": "issue:38175",
+            "right": "issue:43493",
+            "accept": false,
+            "reason": "Both mention SigLIP2, but one is zero probabilities while the other is a HF-vs-JAX discrepancy."
+          },
+          {
+            "left": "issue:44442",
+            "right": "issue:45198",
+            "accept": false,
+            "reason": "Tokenizer load failure and save/tokenization failure are separate model-specific issues."
+          },
+          {
+            "left": "issue:43278",
+            "right": "issue:43856",
+            "accept": false,
+            "reason": "Embedding dtype mismatch in eval is unrelated to MoE training memory usage."
+          },
+          {
+            "left": "issue:43827",
+            "right": "issue:43901",
+            "accept": false,
+            "reason": "Both touch docs/pipeline behavior, but they concern different API topics and fixes."
+          },
+          {
+            "left": "issue:43653",
+            "right": "issue:44112",
+            "accept": false,
+            "reason": "Tokenizer special-token registration bug vs CI test staleness are unrelated."
+          },
+          {
+            "left": "issue:43450",
+            "right": "issue:43493",
+            "accept": false,
+            "reason": "Batched video processor shape bug is unrelated to SigLIP2 model discrepancy."
+          },
+          {
+            "left": "issue:43450",
+            "right": "issue:43575",
+            "accept": false,
+            "reason": "Video preprocessing shape issue and tensor-parallel OOM are not the same bug."
+          },
+          {
+            "left": "issue:44792",
+            "right": "issue:45563",
+            "accept": false,
+            "reason": "Different generation issues: failing Janus test vs stale warning for paged generate()."
+          },
+          {
+            "left": "issue:45127",
+            "right": "issue:45245",
+            "accept": false,
+            "reason": "LoRA merge collapse and category-count runtime error have no shared code path."
+          },
+          {
+            "left": "issue:44360",
+            "right": "issue:44561",
+            "accept": false,
+            "reason": "A missing ReLU in one component is unrelated to removal of a torch-fx helper."
+          },
+          {
+            "left": "issue:43819",
+            "right": "issue:45083",
+            "accept": false,
+            "reason": "Different numerical/helper-function bugs in different models."
+          },
+          {
+            "left": "issue:44206",
+            "right": "issue:45584",
+            "accept": false,
+            "reason": "Unsupported feature-extractor arg and empty-transcription Whisper failure are unrelated."
+          },
+          {
+            "left": "issue:43819",
+            "right": "issue:44112",
+            "accept": false,
+            "reason": "DAC latent mismatch and a stale CI test are unrelated."
+          },
+          {
+            "left": "issue:43262",
+            "right": "issue:43450",
+            "accept": false,
+            "reason": "Audio chat-template sampling-rate default and batched video shape are unrelated."
+          },
+          {
+            "left": "issue:44610",
+            "right": "issue:44857",
+            "accept": false,
+            "reason": "Processor resize mismatch vs AMP/CUDA loss crash are different bugs."
+          },
+          {
+            "left": "issue:37428",
+            "right": "issue:43425",
+            "accept": false,
+            "reason": "ImportError on a missing symbol is unrelated to Torch 2.10 compatibility concerns."
+          },
+          {
+            "left": "issue:44857",
+            "right": "issue:45593",
+            "accept": false,
+            "reason": "Different training bugs: AMP/CUDA crash vs auxiliary-loss omission."
+          },
+          {
+            "left": "issue:43278",
+            "right": "issue:43825",
+            "accept": false,
+            "reason": "Embedding dtype regression is unrelated to pipeline error messaging."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 23,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:28218",
+        "issue:29942",
+        "issue:32090",
+        "issue:35141",
+        "issue:37428",
+        "issue:38175",
+        "issue:39692",
+        "issue:43335",
+        "issue:43388",
+        "issue:43504",
+        "issue:43526",
+        "issue:43550",
+        "issue:43606",
+        "issue:43611",
+        "issue:43653",
+        "issue:43720",
+        "issue:43819",
+        "issue:43931",
+        "issue:44112",
+        "issue:44206",
+        "issue:44292",
+        "issue:44351",
+        "issue:44448",
+        "issue:44462",
+        "issue:44493",
+        "issue:44561",
+        "issue:44568",
+        "issue:44743",
+        "issue:44779",
+        "issue:44855",
+        "issue:44857",
+        "issue:44977",
+        "issue:45072",
+        "issue:45083",
+        "issue:45200",
+        "issue:45229",
+        "issue:45290",
+        "issue:45310",
+        "issue:45381",
+        "issue:45459",
+        "issue:45478",
+        "issue:45491",
+        "issue:45520",
+        "issue:45588",
+        "issue:45593",
+        "issue:45712",
+        "issue:45823",
+        "issue:45859"
+      ],
+      "soft_pairs": [
+        "issue:45290|issue:45381",
+        "issue:43606|issue:45083",
+        "issue:43335|issue:43550",
+        "issue:45200|issue:45588",
+        "issue:44448|issue:44743",
+        "issue:37428|issue:44351",
+        "issue:43611|issue:43931",
+        "issue:39692|issue:43526",
+        "issue:44462|issue:44561",
+        "issue:44206|issue:45823",
+        "issue:44561|issue:44779",
+        "issue:43388|issue:43526",
+        "issue:43335|issue:43653",
+        "issue:43819|issue:44493",
+        "issue:45310|issue:45478",
+        "issue:44855|issue:45083",
+        "issue:45588|issue:45859",
+        "issue:43550|issue:44112",
+        "issue:43335|issue:45083",
+        "issue:29942|issue:32090",
+        "issue:43335|issue:43504",
+        "issue:39692|issue:44493",
+        "issue:44292|issue:45520",
+        "issue:38175|issue:44493",
+        "issue:44448|issue:44779",
+        "issue:28218|issue:35141",
+        "issue:43335|issue:45072",
+        "issue:43335|issue:45593",
+        "issue:43720|issue:44112",
+        "issue:44977|issue:45229",
+        "issue:44448|issue:44977",
+        "issue:44568|issue:45491",
+        "issue:43606|issue:44857",
+        "issue:44857|issue:45072",
+        "issue:45459|issue:45712"
+      ],
+      "prepared_review_unit_hash": "ba8689049f719a6e53eba3d2b8eda9f35a2ef6a03e8d3a8bd06279350cc2cdbb",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 35,
+        "serialized_chars": 38458,
+        "estimated_input_tokens": 9615,
+        "estimated_eval_tokens": 19486
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "Most items are unrelated issue reports; only one soft pair looks like a true duplicate. The two Qwen3.5 Moe from_pretrained reports are the same bug, while the rest are different failure modes or subsystems.",
+        "confidence": 0.95,
+        "canonical_issue_reason": "issue:45310 is the best canonical issue because it is the earlier of the two exact duplicate reports for the same Qwen3.5 Moe from_pretrained regression.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "issue:45310 is the strongest issue candidate overall: it has the exact duplicate match, is earlier, and best represents the confirmed duplicate bug in this set.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:45290",
+            "right": "issue:45381",
+            "accept": false,
+            "reason": "Different bugs: chat template/tool-call crash vs qwen2.5-vl video position-id mismatch."
+          },
+          {
+            "left": "issue:43606",
+            "right": "issue:45083",
+            "accept": false,
+            "reason": "Different subsystems and failures: CPU offload device mismatch vs helper length computation in qwen3_omni_moe."
+          },
+          {
+            "left": "issue:43335",
+            "right": "issue:43550",
+            "accept": false,
+            "reason": "Different concrete problems: SwitchTransformers sparse-layer config bug vs Bamba SDPA torch.compile failure."
+          },
+          {
+            "left": "issue:45200",
+            "right": "issue:45588",
+            "accept": false,
+            "reason": "Different bugs: Gemma4 mm_token_type_ids defaults vs flash_attention s_aux=None crash."
+          },
+          {
+            "left": "issue:44448",
+            "right": "issue:44743",
+            "accept": false,
+            "reason": "Different models and code paths: Pegasus v4/v5 output drift vs qwen3_5 recurrent-state reset."
+          },
+          {
+            "left": "issue:37428",
+            "right": "issue:44351",
+            "accept": false,
+            "reason": "Both are import errors, but for different missing symbols and different code paths."
+          },
+          {
+            "left": "issue:43611",
+            "right": "issue:43931",
+            "accept": false,
+            "reason": "Different loading failures: base_model_prefix handling vs Qwen3-VL weight shape mismatch."
+          },
+          {
+            "left": "issue:39692",
+            "right": "issue:43526",
+            "accept": false,
+            "reason": "Different BEiT/SigLIP2 issues: documentation example errors vs BeitImageProcessorFast reduce_labels bug."
+          },
+          {
+            "left": "issue:44462",
+            "right": "issue:44561",
+            "accept": false,
+            "reason": "Different regressions: AutoTokenizer ignoring tokenizer.json vs removal of is_torch_fx_available breaking trust_remote_code."
+          },
+          {
+            "left": "issue:44206",
+            "right": "issue:45823",
+            "accept": false,
+            "reason": "Different model bugs: unsupported center arg in LasrFeatureExtractor vs Gemma4 PLE device mismatch."
+          },
+          {
+            "left": "issue:44561",
+            "right": "issue:44779",
+            "accept": false,
+            "reason": "Unrelated breakages: trust_remote_code import regression vs Deepseek tokenizer behavior change."
+          },
+          {
+            "left": "issue:43388",
+            "right": "issue:43526",
+            "accept": false,
+            "reason": "Different data handling bugs: gather_for_metrics last-batch label truncation vs BeitImageProcessorFast reduce_labels output."
+          },
+          {
+            "left": "issue:43335",
+            "right": "issue:43653",
+            "accept": false,
+            "reason": "Different issues: SwitchTransformers sparse-layer creation vs BigBirdTokenizer special-token registration."
+          },
+          {
+            "left": "issue:43819",
+            "right": "issue:44493",
+            "accept": false,
+            "reason": "Different bugs: DAC.from_latents mismatch vs unexpected position-id key warnings across models."
+          },
+          {
+            "left": "issue:45310",
+            "right": "issue:45478",
+            "accept": true,
+            "reason": "Exact same Qwen3.5 Moe from_pretrained error; same title and same regression report."
+          },
+          {
+            "left": "issue:44855",
+            "right": "issue:45083",
+            "accept": false,
+            "reason": "Different failures: Python 3.13 import indentation error vs qwen3_omni_moe feature-length helper behavior."
+          },
+          {
+            "left": "issue:45588",
+            "right": "issue:45859",
+            "accept": false,
+            "reason": "Different flash-attention-related bugs: s_aux=None crash vs CPU-offload incompatibility in text rotary embedding."
+          },
+          {
+            "left": "issue:43550",
+            "right": "issue:44112",
+            "accept": false,
+            "reason": "Different problems: Bamba torch.compile/SDPA failure vs GraniteSpeech stale device override test."
+          },
+          {
+            "left": "issue:43335",
+            "right": "issue:45072",
+            "accept": false,
+            "reason": "Different failures: SwitchTransformers sparse-layer config bug vs dtype mismatch in bfloat16 inference."
+          },
+          {
+            "left": "issue:29942",
+            "right": "issue:32090",
+            "accept": false,
+            "reason": "Different issues: Flash Attention 2 tests failing vs Trainer NoneType broadcast error."
+          },
+          {
+            "left": "issue:43335",
+            "right": "issue:43504",
+            "accept": false,
+            "reason": "Different model/config bugs: sparse-layer creation vs BEiT legacy-field loading failure."
+          },
+          {
+            "left": "issue:39692",
+            "right": "issue:44493",
+            "accept": false,
+            "reason": "Different symptom sets: SigLIP2 doc/model mismatch vs unexpected key with position id."
+          },
+          {
+            "left": "issue:44292",
+            "right": "issue:45520",
+            "accept": false,
+            "reason": "Different runtime failures: Qwen-3-8B-NVFP4 error vs flash_attn import KeyError on Python 3.13."
+          },
+          {
+            "left": "issue:38175",
+            "right": "issue:44493",
+            "accept": false,
+            "reason": "Different model behaviors: zero probabilities in SigLIP2 vs unexpected position-id key warnings."
+          },
+          {
+            "left": "issue:44448",
+            "right": "issue:44779",
+            "accept": false,
+            "reason": "Different models and regressions: Pegasus output mismatch vs Deepseek tokenizer breakage in v5."
+          },
+          {
+            "left": "issue:28218",
+            "right": "issue:35141",
+            "accept": false,
+            "reason": "Different tokenizer/model initialization bugs: added-token spacing vs output embedding reinitialization after resizing."
+          },
+          {
+            "left": "issue:43335",
+            "right": "issue:45072",
+            "accept": false,
+            "reason": "Different failures: SwitchTransformers sparse-layer creation vs dtype mismatch during bfloat16 inference."
+          },
+          {
+            "left": "issue:43335",
+            "right": "issue:45593",
+            "accept": false,
+            "reason": "Different model issues: SwitchTransformers config bug vs D-FINE auxiliary-loss handling."
+          },
+          {
+            "left": "issue:43720",
+            "right": "issue:44112",
+            "accept": false,
+            "reason": "Different issues: BitNet packed-weight loading vs GraniteSpeech device override test flakiness."
+          },
+          {
+            "left": "issue:44977",
+            "right": "issue:45229",
+            "accept": false,
+            "reason": "Different bugs: Qwen3.5 flash-attention generation failure vs Gemma4 multi-GPU CUDA OOM."
+          },
+          {
+            "left": "issue:44448",
+            "right": "issue:44977",
+            "accept": false,
+            "reason": "Different models and code paths: Pegasus v4/v5 output difference vs Qwen3.5 flash-attention failure."
+          },
+          {
+            "left": "issue:44568",
+            "right": "issue:45491",
+            "accept": false,
+            "reason": "Different tokenizer/model bugs: mdeberta add_special_tokens regression vs Gemma3 mixed-length NaN embeddings."
+          },
+          {
+            "left": "issue:43606",
+            "right": "issue:44857",
+            "accept": false,
+            "reason": "Different failures: CPU offload device mismatch vs float16 AMP crash in LwDetrImageLoss."
+          },
+          {
+            "left": "issue:44857",
+            "right": "issue:45072",
+            "accept": false,
+            "reason": "Both involve mixed precision, but the models and failure mechanisms are unrelated."
+          },
+          {
+            "left": "issue:45459",
+            "right": "issue:45712",
+            "accept": false,
+            "reason": "Different bugs: tokenizer error masking with protobuf absent vs leftover dummy classes leaking into API surface."
+          }
+        ]
+      },
+      "evaluator_result": {
+        "accept": true,
+        "feedback": "Grounded in the packet: issue:45310 and issue:45478 share the same title and clearly point to the same Qwen3.5 Moe from_pretrained regression, so choosing the earlier report as canonical is reasonable. The other soft pairs are handled conservatively as non-matches based on distinct titles/subsystems."
+      },
+      "evaluator_used": true,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": true,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 24,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:25251",
+        "issue:29127",
+        "issue:29942",
+        "issue:32090",
+        "issue:34689",
+        "issue:35141",
+        "issue:36010",
+        "issue:37428",
+        "issue:40444",
+        "issue:42915",
+        "issue:42994",
+        "issue:43065",
+        "issue:43066",
+        "issue:43278",
+        "issue:43381",
+        "issue:43412",
+        "issue:43421",
+        "issue:43526",
+        "issue:43550",
+        "issue:43653",
+        "issue:43701",
+        "issue:43873",
+        "issue:43927",
+        "issue:43976",
+        "issue:43994",
+        "issue:44186",
+        "issue:44448",
+        "issue:44451",
+        "issue:44462",
+        "issue:44466",
+        "issue:44493",
+        "issue:44743",
+        "issue:44857",
+        "issue:44977",
+        "issue:45083",
+        "issue:45198",
+        "issue:45245",
+        "issue:45372",
+        "issue:45412",
+        "issue:45440",
+        "issue:45459",
+        "issue:45491",
+        "issue:45520",
+        "issue:45563",
+        "issue:45588",
+        "issue:45698",
+        "issue:45803",
+        "issue:45823"
+      ],
+      "soft_pairs": [
+        "issue:29127|issue:32090",
+        "issue:44977|issue:45563",
+        "issue:43927|issue:45698",
+        "issue:45198|issue:45803",
+        "issue:34689|issue:35141",
+        "issue:40444|issue:43526",
+        "issue:43701|issue:45083",
+        "issue:45372|issue:45588",
+        "issue:43550|issue:45083",
+        "issue:43412|issue:45588",
+        "issue:45245|issue:45563",
+        "issue:29942|issue:35141",
+        "issue:43873|issue:44493",
+        "issue:36010|issue:37428",
+        "issue:43066|issue:44462",
+        "issue:45372|issue:45459",
+        "issue:44857|issue:45588",
+        "issue:44448|issue:44466",
+        "issue:43065|issue:45412",
+        "issue:43278|issue:43381",
+        "issue:43653|issue:44743",
+        "issue:45588|issue:45823",
+        "issue:25251|issue:32090",
+        "issue:44451|issue:45520",
+        "issue:43976|issue:45440",
+        "issue:44186|issue:45198",
+        "issue:42915|issue:42994",
+        "issue:43994|issue:45412",
+        "issue:43421|issue:45491"
+      ],
+      "prepared_review_unit_hash": "6a30dfacdda4571cabe2203178f2a1ff8f77005a093002709f8c7459526bb7b8",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 29,
+        "serialized_chars": 36380,
+        "estimated_input_tokens": 9095,
+        "estimated_eval_tokens": 18446
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "The set is mostly a collection of unrelated singleton bugs; none of the soft pairs look like true duplicates or mergeable same-bug reports.",
+        "confidence": 0.94,
+        "canonical_issue_reason": "issue:43873 is the most active, broadly scoped, still-open bug report in the set, so it is the best anchor even though the items are largely unrelated.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "issue:43873 has the strongest combination of breadth, discussion, and ongoing relevance, making it the best representative issue here.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:29127",
+            "right": "issue:32090",
+            "accept": false,
+            "reason": "Different subsystems and failure modes: LayoutLMv3 error messaging vs Trainer GPU broadcast NoneType."
+          },
+          {
+            "left": "issue:44977",
+            "right": "issue:45563",
+            "accept": false,
+            "reason": "Both involve generation, but one is a flash-attention correctness bug and the other is a stale warning in paged generate()."
+          },
+          {
+            "left": "issue:43927",
+            "right": "issue:45698",
+            "accept": false,
+            "reason": "Both are save/load related, but one loses custom token IDs while the other loads the wrong custom module; different code paths."
+          },
+          {
+            "left": "issue:45198",
+            "right": "issue:45803",
+            "accept": false,
+            "reason": "Wav2Vec2 save/tokenization failure is unrelated to Fuyu's broad exception swallowing in tensor conversion."
+          },
+          {
+            "left": "issue:34689",
+            "right": "issue:35141",
+            "accept": false,
+            "reason": "Model loading regression for Llama vision is not the same as post_init reinitializing untied output embeddings."
+          },
+          {
+            "left": "issue:40444",
+            "right": "issue:43526",
+            "accept": false,
+            "reason": "Multiple images per prompt in Qwen2.5-VL and BeitImageProcessorFast label reduction are unrelated bugs."
+          },
+          {
+            "left": "issue:43701",
+            "right": "issue:45083",
+            "accept": false,
+            "reason": "Checkpoint key mismatch is unrelated to qwen3_omni_moe feature-length helper behavior."
+          },
+          {
+            "left": "issue:45372",
+            "right": "issue:45588",
+            "accept": false,
+            "reason": "Gemma4 processor import failure and flash_attention s_aux=None crash are different errors in different components."
+          },
+          {
+            "left": "issue:43550",
+            "right": "issue:45083",
+            "accept": false,
+            "reason": "torch.compile/SDPA failure in Bamba is unrelated to qwen3_omni_moe feature-length computation."
+          },
+          {
+            "left": "issue:43412",
+            "right": "issue:45588",
+            "accept": false,
+            "reason": "ExecuTorch export guard/support issue is unrelated to sink-less flash_attention AttributeError."
+          },
+          {
+            "left": "issue:45245",
+            "right": "issue:45563",
+            "accept": false,
+            "reason": "Category-cardinality runtime error and stale generate warning are not the same bug."
+          },
+          {
+            "left": "issue:29942",
+            "right": "issue:35141",
+            "accept": false,
+            "reason": "Flash Attention 2 test failures do not match embedding reinitialization after resize."
+          },
+          {
+            "left": "issue:43873",
+            "right": "issue:44493",
+            "accept": false,
+            "reason": "Quantization/offloading behavior and unexpected position-id keys are distinct issues."
+          },
+          {
+            "left": "issue:36010",
+            "right": "issue:37428",
+            "accept": false,
+            "reason": "Both are import errors, but they concern different symbols and different modules."
+          },
+          {
+            "left": "issue:43066",
+            "right": "issue:44462",
+            "accept": false,
+            "reason": "Tokenizer decoder type mismatch in v5 is not the same as AutoTokenizer ignoring tokenizer.json."
+          },
+          {
+            "left": "issue:45372",
+            "right": "issue:45459",
+            "accept": false,
+            "reason": "Different tokenizer-related failures: missing ReasoningEffort vs protobuf exception masking."
+          },
+          {
+            "left": "issue:44857",
+            "right": "issue:45588",
+            "accept": false,
+            "reason": "AMP/CUDA loss crash in LwDetr is unrelated to flash_attention sink handling."
+          },
+          {
+            "left": "issue:44448",
+            "right": "issue:44466",
+            "accept": false,
+            "reason": "One is output differences across versions; the other is inconsistent weight serialization by device."
+          },
+          {
+            "left": "issue:43065",
+            "right": "issue:45412",
+            "accept": false,
+            "reason": "Dummy Conv2d in Sam3PixelDecoder and RT-DETR memory reclamation are unrelated."
+          },
+          {
+            "left": "issue:43278",
+            "right": "issue:43381",
+            "accept": false,
+            "reason": "Eval-mode dtype drift and gradient-checkpointing restrictions are different symptoms and fixes."
+          },
+          {
+            "left": "issue:43653",
+            "right": "issue:44743",
+            "accept": false,
+            "reason": "Tokenizer special-token registration and recurrent-state reset with cache are unrelated."
+          },
+          {
+            "left": "issue:45588",
+            "right": "issue:45823",
+            "accept": false,
+            "reason": "flash_attention None-handling crash and Gemma4 device mismatch are different forward-path bugs."
+          },
+          {
+            "left": "issue:25251",
+            "right": "issue:32090",
+            "accept": false,
+            "reason": "Pipeline top_k nesting behavior is unrelated to Trainer GPU broadcast NoneType handling."
+          },
+          {
+            "left": "issue:44451",
+            "right": "issue:45520",
+            "accept": false,
+            "reason": "Model loading failure for ScandiBERT and Python 3.13 flash_attn import KeyError are unrelated."
+          },
+          {
+            "left": "issue:43976",
+            "right": "issue:45440",
+            "accept": false,
+            "reason": "Python-version compatibility issue and DeepSeekV3MoE divergence are not the same problem."
+          },
+          {
+            "left": "issue:44186",
+            "right": "issue:45198",
+            "accept": false,
+            "reason": "LayoutLMv2 tokenizer padding/truncation crash and Wav2Vec2 save/tokenization failure are different tokenizer bugs."
+          },
+          {
+            "left": "issue:42915",
+            "right": "issue:42994",
+            "accept": false,
+            "reason": "Quantization config failure on Qwen3Moe and quantized saving failure are related only broadly, not the same bug."
+          },
+          {
+            "left": "issue:43994",
+            "right": "issue:45412",
+            "accept": false,
+            "reason": "SigLIP2 inference weirdness and RT-DETR memory release are unrelated."
+          },
+          {
+            "left": "issue:43421",
+            "right": "issue:45491",
+            "accept": false,
+            "reason": "Runtime special-token post-processor updates and Gemma3 NaN embeddings are different components and failure modes."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 25,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:29127",
+        "issue:29942",
+        "issue:30064",
+        "issue:32090",
+        "issue:33357",
+        "issue:34567",
+        "issue:36010",
+        "issue:39692",
+        "issue:41720",
+        "issue:43065",
+        "issue:43278",
+        "issue:43329",
+        "issue:43334",
+        "issue:43381",
+        "issue:43526",
+        "issue:43531",
+        "issue:43575",
+        "issue:43653",
+        "issue:43701",
+        "issue:43723",
+        "issue:43756",
+        "issue:43761",
+        "issue:43824",
+        "issue:43873",
+        "issue:43994",
+        "issue:44186",
+        "issue:44206",
+        "issue:44265",
+        "issue:44279",
+        "issue:44361",
+        "issue:44442",
+        "issue:44479",
+        "issue:44554",
+        "issue:44610",
+        "issue:44734",
+        "issue:44871",
+        "issue:44977",
+        "issue:45103",
+        "issue:45198",
+        "issue:45229",
+        "issue:45245",
+        "issue:45372",
+        "issue:45406",
+        "issue:45412",
+        "issue:45440",
+        "issue:45593",
+        "issue:45834",
+        "issue:45859"
+      ],
+      "soft_pairs": [
+        "issue:44361|issue:45198",
+        "issue:44871|issue:44977",
+        "issue:44206|issue:44479",
+        "issue:44265|issue:45198",
+        "issue:44734|issue:45406",
+        "issue:29942|issue:30064",
+        "issue:43065|issue:43994",
+        "issue:33357|issue:39692",
+        "issue:29127|issue:34567",
+        "issue:44610|issue:45412",
+        "issue:45103|issue:45372",
+        "issue:43756|issue:45440",
+        "issue:43723|issue:44554",
+        "issue:43873|issue:45229",
+        "issue:45372|issue:45406",
+        "issue:43278|issue:43575",
+        "issue:43761|issue:45859",
+        "issue:43526|issue:45834",
+        "issue:43381|issue:43701",
+        "issue:43653|issue:44442",
+        "issue:45440|issue:45593",
+        "issue:41720|issue:45229",
+        "issue:43278|issue:43329",
+        "issue:29942|issue:36010",
+        "issue:32090|issue:33357",
+        "issue:43334|issue:43531",
+        "issue:44977|issue:45245",
+        "issue:43526|issue:44186",
+        "issue:43824|issue:44279"
+      ],
+      "prepared_review_unit_hash": "b3241eac8893d05a59e4a3fa0cc7ea0e5622638483496ff8fa8065dd58ad3f61",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 29,
+        "serialized_chars": 36452,
+        "estimated_input_tokens": 9113,
+        "estimated_eval_tokens": 18482
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "The items are a heterogeneous set of unrelated issues, and none of the soft pairs look like true duplicates of the same underlying bug or change. I rejected all soft edges and found no sensible single canonical issue/PR for this cluster.",
+        "confidence": 0.94,
+        "canonical_issue_reason": "No clear canonical issue: the set spans many unrelated subsystems (tokenizers, vision models, trainers, serving, offloading, docs), so there is no single representative duplicate target.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "No strong global best issue for triage; the cluster is too mixed to nominate one issue as the representative duplicate anchor.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:44361",
+            "right": "issue:45198",
+            "accept": false,
+            "reason": "Both are tokenizer-related failures, but they involve different models and different failure modes (`AttributeError` on tasks vs save/tokenization failure), so they are not the same bug."
+          },
+          {
+            "left": "issue:44871",
+            "right": "issue:44977",
+            "accept": false,
+            "reason": "Different models and different problem types: EOS token config mismatch vs flash-attention generation failure."
+          },
+          {
+            "left": "issue:44206",
+            "right": "issue:44479",
+            "accept": false,
+            "reason": "One is an unsupported argument in a feature extractor; the other is a video-input regression across multiple VLMs. Not the same code-path bug."
+          },
+          {
+            "left": "issue:44265",
+            "right": "issue:45198",
+            "accept": false,
+            "reason": "`torch.export.export` with `torch_compilable_check` is unrelated to Wav2Vec2 tokenizer/save_pretrained failures."
+          },
+          {
+            "left": "issue:44734",
+            "right": "issue:45406",
+            "accept": false,
+            "reason": "Serving KV-cache tensor indexing and Gemma4 processor `_tokenizer` initialization are different failures in different code paths."
+          },
+          {
+            "left": "issue:29942",
+            "right": "issue:30064",
+            "accept": false,
+            "reason": "Flash Attention 2 test failures are unrelated to image processor segmentation-map handling."
+          },
+          {
+            "left": "issue:43065",
+            "right": "issue:43994",
+            "accept": false,
+            "reason": "A dummy Conv2d in Sam3PixelDecoder is unrelated to SigLIP2 AutoModel/pipeline correctness."
+          },
+          {
+            "left": "issue:33357",
+            "right": "issue:39692",
+            "accept": false,
+            "reason": "MacOS bus error with a community CLIP model is not the same bug as documentation example model/processor mismatch and quantization errors."
+          },
+          {
+            "left": "issue:29127",
+            "right": "issue:34567",
+            "accept": false,
+            "reason": "LayoutLMv3 error messaging and TrainerState token-count tracking are unrelated issues."
+          },
+          {
+            "left": "issue:44610",
+            "right": "issue:45412",
+            "accept": false,
+            "reason": "Processor image-size mismatch and RT-DETR memory not being released are distinct bugs."
+          },
+          {
+            "left": "issue:45103",
+            "right": "issue:45372",
+            "accept": false,
+            "reason": "Auto-docstring annotation handling and Gemma4 processor loading/import dependencies are unrelated."
+          },
+          {
+            "left": "issue:43756",
+            "right": "issue:45440",
+            "accept": false,
+            "reason": "Both concern model implementation differences, but they affect different models and different mechanisms; not the same concrete bug."
+          },
+          {
+            "left": "issue:43723",
+            "right": "issue:44554",
+            "accept": false,
+            "reason": "Tokenizer loading in v5 and an MPS attention correctness issue are unrelated code paths."
+          },
+          {
+            "left": "issue:43873",
+            "right": "issue:45229",
+            "accept": false,
+            "reason": "Quantization/offloading behavior and multi-GPU CUDA OOM are different failure modes and not the same fix."
+          },
+          {
+            "left": "issue:45372",
+            "right": "issue:45406",
+            "accept": false,
+            "reason": "Both mention Gemma4, but one is an import-time dependency issue and the other is a serving-time processor attribute error."
+          },
+          {
+            "left": "issue:43278",
+            "right": "issue:43575",
+            "accept": false,
+            "reason": "Training/eval dtype drift is unrelated to tensor-parallel load OOM."
+          },
+          {
+            "left": "issue:43761",
+            "right": "issue:45859",
+            "accept": false,
+            "reason": "CLIPVisionModel hidden-state regression and Qwen3_5MoE CPU-offload incompatibility are different model/component bugs."
+          },
+          {
+            "left": "issue:43526",
+            "right": "issue:45834",
+            "accept": false,
+            "reason": "BeitImageProcessorFast label reduction and Kosmos2.5 OCR long-input indexing are unrelated processor bugs."
+          },
+          {
+            "left": "issue:43381",
+            "right": "issue:43701",
+            "accept": false,
+            "reason": "Gradient-checkpointing eval-mode behavior and resume-from-checkpoint key mismatch are not the same underlying issue."
+          },
+          {
+            "left": "issue:43653",
+            "right": "issue:44442",
+            "accept": false,
+            "reason": "BigBirdTokenizer special-token registration and FastSpeech2ConformerTokenizer loading failure are different tokenizer bugs."
+          },
+          {
+            "left": "issue:45440",
+            "right": "issue:45593",
+            "accept": false,
+            "reason": "Different models and different implementation gaps: DeepseekV3MoE divergence vs missing D-FINE auxiliary losses."
+          },
+          {
+            "left": "issue:41720",
+            "right": "issue:45229",
+            "accept": false,
+            "reason": "A device-mapping cuda assert and a multi-GPU OOM are distinct runtime failures."
+          },
+          {
+            "left": "issue:43278",
+            "right": "issue:43329",
+            "accept": false,
+            "reason": "Embedding dtype changes and multimodal token counting/video-branch undefined symbols are unrelated."
+          },
+          {
+            "left": "issue:29942",
+            "right": "issue:36010",
+            "accept": false,
+            "reason": "Flash Attention 2 test regressions are unrelated to an import error for `GenerationMixin`."
+          },
+          {
+            "left": "issue:32090",
+            "right": "issue:33357",
+            "accept": false,
+            "reason": "Trainer broadcast TypeError and MacOS bus error in CLIP are unrelated crashes."
+          },
+          {
+            "left": "issue:43334",
+            "right": "issue:43531",
+            "accept": false,
+            "reason": "Qwen3-VL loading failure due to `pad_token_id` and Qwen3-MoE sliding-window behavior are different model-specific bugs."
+          },
+          {
+            "left": "issue:44977",
+            "right": "issue:45245",
+            "accept": false,
+            "reason": "Flash-attention generation failure and category-count limit error are unrelated."
+          },
+          {
+            "left": "issue:43526",
+            "right": "issue:44186",
+            "accept": false,
+            "reason": "Beit label reduction and LayoutLMv2 tokenizer padding/truncation crashes are different components and failures."
+          },
+          {
+            "left": "issue:43824",
+            "right": "issue:44279",
+            "accept": false,
+            "reason": "A specific model import error is not the same as a vague dependency issue with transformers."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 26,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:28218",
+        "issue:29127",
+        "issue:32090",
+        "issue:33290",
+        "issue:33357",
+        "issue:34567",
+        "issue:34689",
+        "issue:35141",
+        "issue:38175",
+        "issue:39692",
+        "issue:42907",
+        "issue:42915",
+        "issue:43262",
+        "issue:43316",
+        "issue:43381",
+        "issue:43441",
+        "issue:43525",
+        "issue:43526",
+        "issue:43575",
+        "issue:43611",
+        "issue:43646",
+        "issue:43701",
+        "issue:43761",
+        "issue:43873",
+        "issue:43931",
+        "issue:44077",
+        "issue:44079",
+        "issue:44188",
+        "issue:44206",
+        "issue:44387",
+        "issue:44442",
+        "issue:44464",
+        "issue:44479",
+        "issue:44743",
+        "issue:44811",
+        "issue:44857",
+        "issue:44898",
+        "issue:44964",
+        "issue:44977",
+        "issue:45072",
+        "issue:45216",
+        "issue:45290",
+        "issue:45440",
+        "issue:45491",
+        "issue:45584",
+        "issue:45588",
+        "issue:45698",
+        "issue:45803"
+      ],
+      "soft_pairs": [
+        "issue:43526|issue:44857",
+        "issue:43262|issue:44206",
+        "issue:43441|issue:45588",
+        "issue:44464|issue:45491",
+        "issue:42915|issue:43931",
+        "issue:44743|issue:44811",
+        "issue:44442|issue:45072",
+        "issue:43575|issue:43701",
+        "issue:43646|issue:44077",
+        "issue:29127|issue:35141",
+        "issue:43575|issue:44387",
+        "issue:44442|issue:45803",
+        "issue:44977|issue:45584",
+        "issue:33357|issue:38175",
+        "issue:43316|issue:43525",
+        "issue:44964|issue:45440",
+        "issue:44479|issue:45290",
+        "issue:43761|issue:44079",
+        "issue:28218|issue:29127",
+        "issue:42907|issue:45698",
+        "issue:28218|issue:32090",
+        "issue:43931|issue:45216",
+        "issue:33290|issue:34567",
+        "issue:34689|issue:43611",
+        "issue:39692|issue:45491",
+        "issue:44442|issue:44898",
+        "issue:43761|issue:44188",
+        "issue:43381|issue:43873"
+      ],
+      "prepared_review_unit_hash": "ca3e7191d5861bc6959c4833b3f5e20f18d83b5a019c000e32bee11e7fd43f47",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 28,
+        "serialized_chars": 36130,
+        "estimated_input_tokens": 9033,
+        "estimated_eval_tokens": 18322
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "These items are heterogeneous and the proposed soft pairs do not appear to be the same underlying bug or change. Most only share a broad subsystem name (e.g. flash attention, loading/saving, OOM) but differ in concrete code path and symptom.",
+        "confidence": 0.93,
+        "canonical_issue_reason": "No single canonical issue stands out because the cluster is not a true duplicate set; the items span unrelated bugs across tokenization, processors, model loading, quantization, training, and serialization.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "No global best issue is suitable: none of these issues represent the shared problem across the set, so picking one would be arbitrary.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:43526",
+            "right": "issue:44857",
+            "accept": false,
+            "reason": "Different failures: label reduction bug in BEiT fast processor vs AMP/CUDA crash in image loss."
+          },
+          {
+            "left": "issue:43262",
+            "right": "issue:44206",
+            "accept": false,
+            "reason": "Audio chat-template sampling-rate default mismatch is unrelated to LasrFeatureExtractor's unsupported argument crash."
+          },
+          {
+            "left": "issue:43441",
+            "right": "issue:45588",
+            "accept": false,
+            "reason": "Both involve flash attention, but one is a Ministral-3 v5 RC regression and the other is a sink-less model AttributeError on s_aux=None."
+          },
+          {
+            "left": "issue:44464",
+            "right": "issue:45491",
+            "accept": false,
+            "reason": "Compiled-forward chunked generation inconsistency is a different bug from Gemma3 NaN embeddings on mixed-length sliding-window batches."
+          },
+          {
+            "left": "issue:42915",
+            "right": "issue:43931",
+            "accept": false,
+            "reason": "Qwen3Moe FP8Config failure and Qwen3-VL weight-shape mismatch are different model/load paths with different symptoms."
+          },
+          {
+            "left": "issue:44743",
+            "right": "issue:44811",
+            "accept": false,
+            "reason": "Recurrent-state reset in modular_qwen3_5 is unrelated to Whisper batch_decode ignoring skip_special_tokens."
+          },
+          {
+            "left": "issue:44442",
+            "right": "issue:45072",
+            "accept": false,
+            "reason": "Tokenizer load failure and dtype mismatches in inference are unrelated bugs."
+          },
+          {
+            "left": "issue:43575",
+            "right": "issue:43701",
+            "accept": false,
+            "reason": "Tensor-parallel OOM while loading a model is not the same as a resume_from_checkpoint key mismatch."
+          },
+          {
+            "left": "issue:43646",
+            "right": "issue:44077",
+            "accept": false,
+            "reason": "Broad custom initialization regression vs a specific patchtsmixer post_init policy issue; not the same bug."
+          },
+          {
+            "left": "issue:29127",
+            "right": "issue:35141",
+            "accept": false,
+            "reason": "LayoutLMv3 error-message clarity issue is unrelated to embedding reinitialization after resize_token_embeddings."
+          },
+          {
+            "left": "issue:43575",
+            "right": "issue:44387",
+            "accept": false,
+            "reason": "Both are OOM-related, but one is TP load OOM and the other is increased CUDA reserved memory under int4 quantization."
+          },
+          {
+            "left": "issue:44442",
+            "right": "issue:45803",
+            "accept": false,
+            "reason": "AutoTokenizer load failure is unrelated to FuyuBatchFeature's bare except swallowing interrupts/errors."
+          },
+          {
+            "left": "issue:44977",
+            "right": "issue:45584",
+            "accept": false,
+            "reason": "Qwen3.5 flash-attention generation issue and Whisper empty-transcription failure after align_special_tokens are different code paths."
+          },
+          {
+            "left": "issue:33357",
+            "right": "issue:38175",
+            "accept": false,
+            "reason": "MacOS bus error with CLIP model loading is not the same as SigLIP2 zero-probability outputs."
+          },
+          {
+            "left": "issue:43316",
+            "right": "issue:43525",
+            "accept": false,
+            "reason": "Gemma3TextConfig API discrepancy and Llama4Config missing pad_token_id are distinct config-contract bugs."
+          },
+          {
+            "left": "issue:44964",
+            "right": "issue:45440",
+            "accept": false,
+            "reason": "Phi-4 multimodal loading failure is not the same as DeepseekV3MoE diverging from the remote implementation."
+          },
+          {
+            "left": "issue:44479",
+            "right": "issue:45290",
+            "accept": false,
+            "reason": "Video input regression across Qwen VL models is unrelated to apply_chat_template crashing on tool-call assistant messages with no content."
+          },
+          {
+            "left": "issue:43761",
+            "right": "issue:44079",
+            "accept": false,
+            "reason": "CLIPVisionModel hidden_states regression is unrelated to ModelOutput key assignment when prior values were None."
+          },
+          {
+            "left": "issue:28218",
+            "right": "issue:29127",
+            "accept": false,
+            "reason": "Tokenizer spacing after added token is unrelated to layoutlmv3 error-message clarity."
+          },
+          {
+            "left": "issue:42907",
+            "right": "issue:45698",
+            "accept": false,
+            "reason": "Saving dequantized Ministral/Devstral models and loading the wrong custom module after save_pretrained are different serialization failures."
+          },
+          {
+            "left": "issue:28218",
+            "right": "issue:32090",
+            "accept": false,
+            "reason": "Tokenizer extra-space behavior has no relation to Trainer _gpu_broadcast_one NoneType TypeError."
+          },
+          {
+            "left": "issue:43931",
+            "right": "issue:45216",
+            "accept": false,
+            "reason": "Model weight-shape mismatch on load is different from Qwen3.5 checkpoint corruption after save_pretrained."
+          },
+          {
+            "left": "issue:33290",
+            "right": "issue:34567",
+            "accept": false,
+            "reason": "Adafactor OOM in deepspeed is unrelated to TrainerState.num_input_tokens_seen not updating."
+          },
+          {
+            "left": "issue:34689",
+            "right": "issue:43611",
+            "accept": false,
+            "reason": "Llama 3.2 Vision loading regression and base_model_prefix loading breakage are both load-related but not the same concrete defect."
+          },
+          {
+            "left": "issue:39692",
+            "right": "issue:45491",
+            "accept": false,
+            "reason": "SigLIP2 documentation example errors are unrelated to Gemma3 NaN embeddings during mixed-length batching."
+          },
+          {
+            "left": "issue:44442",
+            "right": "issue:44898",
+            "accept": false,
+            "reason": "Tokenizer loading failure is unrelated to Perceiver non-default resolution failure despite both being model/processor issues."
+          },
+          {
+            "left": "issue:43761",
+            "right": "issue:44188",
+            "accept": false,
+            "reason": "Hidden-states regression in CLIPVisionModel is unrelated to attention-kernel divergence under torch.compile."
+          },
+          {
+            "left": "issue:43381",
+            "right": "issue:43873",
+            "accept": false,
+            "reason": "Gradient checkpointing eval-mode restriction is unrelated to offloading behavior under quantization."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 27,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:25251",
+        "issue:28282",
+        "issue:29127",
+        "issue:32090",
+        "issue:33290",
+        "issue:33357",
+        "issue:34689",
+        "issue:36010",
+        "issue:41628",
+        "issue:41950",
+        "issue:42175",
+        "issue:43232",
+        "issue:43295",
+        "issue:43316",
+        "issue:43381",
+        "issue:43388",
+        "issue:43404",
+        "issue:43441",
+        "issue:43475",
+        "issue:43493",
+        "issue:43540",
+        "issue:43550",
+        "issue:43643",
+        "issue:43720",
+        "issue:43742",
+        "issue:43825",
+        "issue:43827",
+        "issue:43856",
+        "issue:43873",
+        "issue:44077",
+        "issue:44162",
+        "issue:44190",
+        "issue:44442",
+        "issue:44479",
+        "issue:44496",
+        "issue:44514",
+        "issue:44568",
+        "issue:44743",
+        "issue:44792",
+        "issue:44855",
+        "issue:44871",
+        "issue:44877",
+        "issue:44913",
+        "issue:45137",
+        "issue:45372",
+        "issue:45479",
+        "issue:45656",
+        "issue:45759"
+      ],
+      "soft_pairs": [
+        "issue:43550|issue:44743",
+        "issue:43493|issue:43827",
+        "issue:44568|issue:45479",
+        "issue:43316|issue:45759",
+        "issue:41628|issue:45372",
+        "issue:43720|issue:44442",
+        "issue:43643|issue:44913",
+        "issue:32090|issue:33290",
+        "issue:44190|issue:44496",
+        "issue:43873|issue:44871",
+        "issue:34689|issue:36010",
+        "issue:43381|issue:43856",
+        "issue:44855|issue:45137",
+        "issue:44077|issue:44479",
+        "issue:33290|issue:45656",
+        "issue:29127|issue:33357",
+        "issue:43316|issue:44877",
+        "issue:43540|issue:44514",
+        "issue:43381|issue:43742",
+        "issue:42175|issue:43316",
+        "issue:33290|issue:34689",
+        "issue:44162|issue:44792",
+        "issue:43404|issue:43441",
+        "issue:25251|issue:28282",
+        "issue:43232|issue:43388",
+        "issue:41950|issue:43475",
+        "issue:43295|issue:43825"
+      ],
+      "prepared_review_unit_hash": "745476ddb7da07127ec60f30a78904010a3494cbf3de5757bfe7501ec902afc7",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 27,
+        "serialized_chars": 36089,
+        "estimated_input_tokens": 9023,
+        "estimated_eval_tokens": 18302
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "This cluster is mostly heterogeneous transformer bugs/regressions, not a tight duplicate set. The soft-similarity pairs are largely false positives across different subsystems and failure modes.",
+        "confidence": 0.39,
+        "canonical_issue_reason": "No true duplicate cluster emerges here; if one issue must anchor the set, issue 41950 is the most generally representative user-facing regression around pipeline/processor handling, but it is not a strong canonical duplicate target.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "Issue 41950 is the best representative issue because it is a concrete, current regression with a clear behavioral failure and broad user impact; however, it still does not subsume the rest of the cluster.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:43550",
+            "right": "issue:44743",
+            "accept": false,
+            "reason": "Different models and failures: torch.compile/SDPA vs cache/state-reset logic."
+          },
+          {
+            "left": "issue:43493",
+            "right": "issue:43827",
+            "accept": false,
+            "reason": "One is a model-implementation discrepancy; the other is documentation/pipeline deprecation cleanup."
+          },
+          {
+            "left": "issue:44568",
+            "right": "issue:45479",
+            "accept": false,
+            "reason": "Tokenizer special-token regression is unrelated to sequence-classification zero-loss behavior."
+          },
+          {
+            "left": "issue:43316",
+            "right": "issue:45759",
+            "accept": false,
+            "reason": "Both concern config/model loading, but for different composite-model API shapes and different bugs."
+          },
+          {
+            "left": "issue:41628",
+            "right": "issue:45372",
+            "accept": false,
+            "reason": "Both are import-related, but they involve different missing symbols and different processors/models."
+          },
+          {
+            "left": "issue:43720",
+            "right": "issue:44442",
+            "accept": false,
+            "reason": "BitNet packed-weight loading and tokenizer loading are unrelated code paths."
+          },
+          {
+            "left": "issue:43643",
+            "right": "issue:44913",
+            "accept": false,
+            "reason": "One is missing fields on trust_remote_code config load; the other is config value loss on reload."
+          },
+          {
+            "left": "issue:32090",
+            "right": "issue:33290",
+            "accept": false,
+            "reason": "Trainer broadcast NoneType error and DeepSpeed Adafactor OOM are different runtime issues."
+          },
+          {
+            "left": "issue:44190",
+            "right": "issue:44496",
+            "accept": false,
+            "reason": "Dataset loading in a script vs model-type/config recognition are separate failures."
+          },
+          {
+            "left": "issue:43873",
+            "right": "issue:44871",
+            "accept": false,
+            "reason": "Quantization offloading behavior and Gemma eos_token_id mismatch are unrelated."
+          },
+          {
+            "left": "issue:34689",
+            "right": "issue:36010",
+            "accept": false,
+            "reason": "Model-loading regression for a vision model and a GenerationMixin import error are different breakages."
+          },
+          {
+            "left": "issue:43381",
+            "right": "issue:43856",
+            "accept": false,
+            "reason": "Gradient checkpointing eval-mode behavior is unrelated to MoE memory usage during training."
+          },
+          {
+            "left": "issue:44855",
+            "right": "issue:45137",
+            "accept": false,
+            "reason": "Python 3.13 torch.jit parse/import issue is unrelated to DeepSpeed ZeRO3 deque underflow."
+          },
+          {
+            "left": "issue:44077",
+            "right": "issue:44479",
+            "accept": false,
+            "reason": "PatchTSMixer post_init API policy and video-input regression for Qwen models are different changes."
+          },
+          {
+            "left": "issue:33290",
+            "right": "issue:45656",
+            "accept": false,
+            "reason": "Both mention optimizer/DeepSpeed, but one is OOM while the other is double optimizer stepping."
+          },
+          {
+            "left": "issue:29127",
+            "right": "issue:33357",
+            "accept": false,
+            "reason": "LayoutLMv3 box-content error messaging and MacOS CLIP bus error are unrelated."
+          },
+          {
+            "left": "issue:43316",
+            "right": "issue:44877",
+            "accept": false,
+            "reason": "Gemma3TextConfig API mismatch is not the same as strict config blocking granite_speech loading."
+          },
+          {
+            "left": "issue:43540",
+            "right": "issue:44514",
+            "accept": false,
+            "reason": "Qwen3OmniMoE video processing ValueError and Qwen2_5_VL batched chat-template crash are distinct paths."
+          },
+          {
+            "left": "issue:43381",
+            "right": "issue:43742",
+            "accept": false,
+            "reason": "Gradient-checkpointing eval behavior does not match a MobileLLM config load key error."
+          },
+          {
+            "left": "issue:42175",
+            "right": "issue:43316",
+            "accept": false,
+            "reason": "Packaging/backends issue is unrelated to Gemma3TextConfig API shape."
+          },
+          {
+            "left": "issue:33290",
+            "right": "issue:34689",
+            "accept": false,
+            "reason": "DeepSpeed Adafactor OOM and Llama 3.2 Vision loading regression are unrelated."
+          },
+          {
+            "left": "issue:44162",
+            "right": "issue:44792",
+            "accept": false,
+            "reason": "ESM2 model breakage and janus test failure are different artifacts and failure modes."
+          },
+          {
+            "left": "issue:43404",
+            "right": "issue:43441",
+            "accept": false,
+            "reason": "Mistral3 weight tying bug and Ministral FlashAttention failure are separate model-specific regressions."
+          },
+          {
+            "left": "issue:25251",
+            "right": "issue:28282",
+            "accept": false,
+            "reason": "Pipeline top_k nested-list behavior is unrelated to PyTorch import absence."
+          },
+          {
+            "left": "issue:43232",
+            "right": "issue:43388",
+            "accept": false,
+            "reason": "Generation kwargs handling after sync_gpus and metric gathering label truncation are different code paths."
+          },
+          {
+            "left": "issue:41950",
+            "right": "issue:43475",
+            "accept": false,
+            "reason": "Video-classification processor selection and SAM3 video encoder attribute error are different bugs."
+          },
+          {
+            "left": "issue:43295",
+            "right": "issue:43825",
+            "accept": false,
+            "reason": "Processor/tokenizer regression is unrelated to pipeline task-support messaging."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 28,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:33290",
+        "issue:35141",
+        "issue:36331",
+        "issue:41720",
+        "issue:42617",
+        "issue:43065",
+        "issue:43295",
+        "issue:43452",
+        "issue:43550",
+        "issue:43575",
+        "issue:43606",
+        "issue:43646",
+        "issue:43673",
+        "issue:43701",
+        "issue:43742",
+        "issue:43761",
+        "issue:43827",
+        "issue:43873",
+        "issue:43906",
+        "issue:43931",
+        "issue:43994",
+        "issue:44060",
+        "issue:44077",
+        "issue:44190",
+        "issue:44242",
+        "issue:44464",
+        "issue:44479",
+        "issue:44496",
+        "issue:44534",
+        "issue:44610",
+        "issue:44617",
+        "issue:44734",
+        "issue:44743",
+        "issue:44792",
+        "issue:44855",
+        "issue:44871",
+        "issue:44913",
+        "issue:44936",
+        "issue:44964",
+        "issue:44977",
+        "issue:45127",
+        "issue:45200",
+        "issue:45245",
+        "issue:45290",
+        "issue:45584",
+        "issue:45588",
+        "issue:45600",
+        "issue:45698"
+      ],
+      "soft_pairs": [
+        "issue:45127|issue:45200",
+        "issue:44871|issue:45584",
+        "issue:43550|issue:43701",
+        "issue:43606|issue:43701",
+        "issue:43295|issue:44871",
+        "issue:43065|issue:44617",
+        "issue:43761|issue:44479",
+        "issue:33290|issue:35141",
+        "issue:33290|issue:36331",
+        "issue:43452|issue:44496",
+        "issue:43673|issue:44743",
+        "issue:43575|issue:43742",
+        "issue:43646|issue:44913",
+        "issue:44936|issue:45698",
+        "issue:42617|issue:43575",
+        "issue:43761|issue:44242",
+        "issue:43575|issue:43873",
+        "issue:44077|issue:44534",
+        "issue:43931|issue:44060",
+        "issue:44464|issue:44871",
+        "issue:44479|issue:44734",
+        "issue:43906|issue:45290",
+        "issue:44617|issue:45584",
+        "issue:44792|issue:44977",
+        "issue:43065|issue:44610",
+        "issue:43827|issue:43994",
+        "issue:44534|issue:44913",
+        "issue:41720|issue:45245",
+        "issue:44190|issue:44964",
+        "issue:44792|issue:45600",
+        "issue:44855|issue:45588"
+      ],
+      "prepared_review_unit_hash": "18db44438cc98f47617e69d2dc5701a0eda09f813bc8cc575dfdc14c0dacde61",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 31,
+        "serialized_chars": 36971,
+        "estimated_input_tokens": 9243,
+        "estimated_eval_tokens": 18742
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "This cluster is not a duplicate set; it mixes unrelated bugs across loading, generation, trainer, quantization, and model-specific regressions. All suggested soft edges look like superficial theme matches rather than the same underlying issue.",
+        "confidence": 0.97,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": null,
+        "best_issue_reason": null,
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:45127",
+            "right": "issue:45200",
+            "accept": false,
+            "reason": "Both involve model/config behavior, but one is LoRA vocabulary merging on tied embeddings and the other is Gemma 4 multimodal token-type defaults; different code paths and failures."
+          },
+          {
+            "left": "issue:44871",
+            "right": "issue:45584",
+            "accept": false,
+            "reason": "Both mention token/eos alignment, but one is a Gemma config inconsistency and the other is Whisper empty-transcription generation after special-token alignment; not the same bug."
+          },
+          {
+            "left": "issue:43550",
+            "right": "issue:43701",
+            "accept": false,
+            "reason": "Torch.compile SDPA failure in Bamba is unrelated to a resume_from_checkpoint key mismatch; no shared underlying code-path."
+          },
+          {
+            "left": "issue:43606",
+            "right": "issue:43701",
+            "accept": false,
+            "reason": "CPU offload device-mismatch in bark-small is a runtime placement bug, while resume_from_checkpoint key mismatch is a checkpoint/key-loading issue."
+          },
+          {
+            "left": "issue:43295",
+            "right": "issue:44871",
+            "accept": false,
+            "reason": "v4.57.5 processor.tokenizer regression and Gemma eos_token_id inconsistency are both config-ish, but they are different regressions in different components."
+          },
+          {
+            "left": "issue:43065",
+            "right": "issue:44617",
+            "accept": false,
+            "reason": "Sam3PixelDecoder dummy Conv2d and Sam3Video CUDA OOM are both Sam3-related, but one is architecture/code correctness and the other is memory usage; not the same bug."
+          },
+          {
+            "left": "issue:43761",
+            "right": "issue:44479",
+            "accept": false,
+            "reason": "CLIPVisionModel hidden_states regression and Qwen video-input regression affect different models and different forward paths."
+          },
+          {
+            "left": "issue:33290",
+            "right": "issue:35141",
+            "accept": false,
+            "reason": "Adafactor+DeepSpeed OOM and embedding resize/post_init reinitialization are unrelated training vs model-initialization problems."
+          },
+          {
+            "left": "issue:33290",
+            "right": "issue:36331",
+            "accept": false,
+            "reason": "OOM in DeepSpeed with Adafactor is unrelated to CustomTrainer.compute_loss receiving an unexpected keyword argument."
+          },
+          {
+            "left": "issue:43452",
+            "right": "issue:44496",
+            "accept": false,
+            "reason": "Both are model-loading failures, but one is gguf_file breaking specific AutoTokenizer/AutoModel loading and the other is an unrecognized model/config.json problem; different failure modes."
+          },
+          {
+            "left": "issue:43673",
+            "right": "issue:44743",
+            "accept": false,
+            "reason": "Both involve cache/state handling, but one is a missing GenerationMixin cache in chunked_prefill and the other is recurrent state reset in modular_qwen3_5; not mergeable as one fix."
+          },
+          {
+            "left": "issue:43575",
+            "right": "issue:43742",
+            "accept": false,
+            "reason": "Qwen2-57B-A14B-Instruct TP OOM and MobileLLM-125M key error are different loading failures with different root causes."
+          },
+          {
+            "left": "issue:43646",
+            "right": "issue:44913",
+            "accept": false,
+            "reason": "Custom model initialization breakage in Transformers 5.0.0 is not the same as GPTNeoXConfig rotary_pct not persisting on reload."
+          },
+          {
+            "left": "issue:44936",
+            "right": "issue:45698",
+            "accept": false,
+            "reason": "trainer.evaluate() after train and wrong custom module loaded after save_pretrained are unrelated training-loop vs serialization/loading bugs."
+          },
+          {
+            "left": "issue:42617",
+            "right": "issue:43575",
+            "accept": false,
+            "reason": "3d_parallel.py execution failure and Qwen2 TP OOM are unrelated; only the broad theme of model scaling overlaps."
+          },
+          {
+            "left": "issue:43761",
+            "right": "issue:44242",
+            "accept": false,
+            "reason": "CLIPVisionModel hidden_states regression is unrelated to load balancing loss not being added when output_router_logits=False."
+          },
+          {
+            "left": "issue:43575",
+            "right": "issue:43873",
+            "accept": false,
+            "reason": "A TP loading OOM and quantization offloading not working are both memory-adjacent, but they are distinct problems in different mechanisms."
+          },
+          {
+            "left": "issue:44077",
+            "right": "issue:44534",
+            "accept": false,
+            "reason": "patchtsmixer post_init allowance and non-persistent buffers filled with junk are both v5 initialization issues, but they are not the same concrete bug."
+          },
+          {
+            "left": "issue:43931",
+            "right": "issue:44060",
+            "accept": false,
+            "reason": "Weight-shape mismatch when loading Qwen3-VL and incorrect tied-weights warning in Qwen3-Next involve different model families and different config logic."
+          },
+          {
+            "left": "issue:44464",
+            "right": "issue:44871",
+            "accept": false,
+            "reason": "Chunked generation inconsistency under compiled forward is unrelated to Gemma eos_token_id configuration."
+          },
+          {
+            "left": "issue:44479",
+            "right": "issue:44734",
+            "accept": false,
+            "reason": "Video-input regression for Qwen multimodal models and /v1/responses KV-cache continuation crash are separate runtime paths."
+          },
+          {
+            "left": "issue:43906",
+            "right": "issue:45290",
+            "accept": false,
+            "reason": "An isolated reproduction of issue 38071 is too generic to match the specific apply_chat_template crash on assistant tool-call messages with no content."
+          },
+          {
+            "left": "issue:44617",
+            "right": "issue:45584",
+            "accept": false,
+            "reason": "Sam3Video CUDA OOM and Whisper empty transcription after align_special_tokens are unrelated model-specific failures."
+          },
+          {
+            "left": "issue:44792",
+            "right": "issue:44977",
+            "accept": false,
+            "reason": "A failing Janus image-generation test and Qwen3.5 flash-attention generation failure are different models and likely different regressions."
+          },
+          {
+            "left": "issue:43065",
+            "right": "issue:44610",
+            "accept": false,
+            "reason": "Sam3PixelDecoder dummy layer and OmDet-Turbo processor producing 640px instead of 224px are different preprocessing/model-shape issues."
+          },
+          {
+            "left": "issue:43827",
+            "right": "issue:43994",
+            "accept": false,
+            "reason": "Docs still referencing pipeline() and SigLIP2 producing nonsensical outputs with AutoModel/pipeline are not the same bug; one is documentation, the other is model behavior."
+          },
+          {
+            "left": "issue:44534",
+            "right": "issue:44913",
+            "accept": false,
+            "reason": "Non-persistent buffer junk and rotary_pct not persisting on reload are both persistence-related, but they affect different configuration/state mechanisms."
+          },
+          {
+            "left": "issue:41720",
+            "right": "issue:45245",
+            "accept": false,
+            "reason": "Qwen3 auto device mapping cudaErrorAssert and category-count ceiling are unrelated runtime vs data-shape constraints."
+          },
+          {
+            "left": "issue:44190",
+            "right": "issue:44964",
+            "accept": false,
+            "reason": "Local dataset loading in a no-trainer example and loading Phi-4 multimodal instruct are separate loading failures."
+          },
+          {
+            "left": "issue:44792",
+            "right": "issue:45600",
+            "accept": false,
+            "reason": "Janus image-generation test failure and stale auto_mappings references to removed Sam3LiteText configs are unrelated CI/model issues."
+          },
+          {
+            "left": "issue:44855",
+            "right": "issue:45588",
+            "accept": false,
+            "reason": "Python 3.13 torch.jit parsing/IndentationError in DebertaV2Model is unrelated to flash_attention crashing on s_aux=None."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 29,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:36246",
+        "issue:38175",
+        "issue:41628",
+        "issue:43065",
+        "issue:43116",
+        "issue:43295",
+        "issue:43335",
+        "issue:43404",
+        "issue:43479",
+        "issue:43540",
+        "issue:43550",
+        "issue:43644",
+        "issue:43650",
+        "issue:43976",
+        "issue:44077",
+        "issue:44112",
+        "issue:44186",
+        "issue:44190",
+        "issue:44220",
+        "issue:44464",
+        "issue:44466",
+        "issue:44496",
+        "issue:44521",
+        "issue:44561",
+        "issue:44683",
+        "issue:44734",
+        "issue:44737",
+        "issue:44855",
+        "issue:44877",
+        "issue:44913",
+        "issue:44933",
+        "issue:44938",
+        "issue:45127",
+        "issue:45216",
+        "issue:45245",
+        "issue:45276",
+        "issue:45357",
+        "issue:45362",
+        "issue:45375",
+        "issue:45381",
+        "issue:45446",
+        "issue:45542",
+        "issue:45563",
+        "issue:45584",
+        "issue:45600",
+        "issue:45632",
+        "issue:45803"
+      ],
+      "soft_pairs": [
+        "issue:44737|issue:45803",
+        "issue:44521|issue:45381",
+        "issue:45127|issue:45357",
+        "issue:43335|issue:44855",
+        "issue:44938|issue:45542",
+        "issue:43976|issue:44855",
+        "issue:43550|issue:44190",
+        "issue:43644|issue:44913",
+        "issue:44077|issue:44877",
+        "issue:44496|issue:45803",
+        "issue:43540|issue:44734",
+        "issue:45276|issue:45584",
+        "issue:44112|issue:45600",
+        "issue:43650|issue:44220",
+        "issue:43335|issue:44190",
+        "issue:43116|issue:44190",
+        "issue:43295|issue:45276",
+        "issue:45245|issue:45584",
+        "issue:43479|issue:45375",
+        "issue:36246|issue:41628",
+        "issue:44464|issue:45563",
+        "issue:44933|issue:45362",
+        "issue:44186|issue:44734",
+        "issue:43404|issue:45216",
+        "issue:44683|issue:45446",
+        "issue:38175|issue:43065",
+        "issue:44734|issue:45632",
+        "issue:44561|issue:45632",
+        "issue:44466|issue:44913"
+      ],
+      "prepared_review_unit_hash": "bcd0a56d1874d66dd55bdda76bd36546446118e34e22fa8b5c36ef035f240540",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 47,
+        "item_count": 47,
+        "soft_pair_count": 29,
+        "serialized_chars": 35958,
+        "estimated_input_tokens": 8990,
+        "estimated_eval_tokens": 18236
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "This cluster is a grab-bag of unrelated Transformers issues, mostly import/config/regression bugs. I found no soft-edge pair that looks like the same underlying bug or change, so no duplicates should be merged here.",
+        "confidence": 0.93,
+        "canonical_issue_reason": "Selected issue 44521 as the best representative because it is a concrete, user-facing API regression in a common multimodal path and has the strongest discussion/reference signal among these otherwise unrelated issues.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "Issue 44521 is the most broadly useful anchor for the cluster: it has a clear repro, affects multimodal chat template behavior, and is more representative of a widely used code path than the more niche import/config bugs.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:44737",
+            "right": "issue:45803",
+            "accept": false,
+            "reason": "Different bugs in different code paths: XLNet positional encoding device placement vs. Fuyu tensor conversion exception handling."
+          },
+          {
+            "left": "issue:44521",
+            "right": "issue:45381",
+            "accept": false,
+            "reason": "Both are multimodal, but one breaks assistant masks in chat templates while the other is a Qwen2.5-VL video position-id issue; not the same defect."
+          },
+          {
+            "left": "issue:45127",
+            "right": "issue:45357",
+            "accept": false,
+            "reason": "Both involve save/weight handling, but they affect different model families and different serialization failures."
+          },
+          {
+            "left": "issue:43335",
+            "right": "issue:44855",
+            "accept": false,
+            "reason": "Unrelated: SwitchTransformers sparse-layer config bug vs. DebertaV2 import failure on Python 3.13."
+          },
+          {
+            "left": "issue:44938",
+            "right": "issue:45542",
+            "accept": false,
+            "reason": "Python 3.14 import/load failure is not the same as a TensorBoard-without-TF backend error."
+          },
+          {
+            "left": "issue:43976",
+            "right": "issue:44855",
+            "accept": false,
+            "reason": "Both mention Python compatibility, but they are different breakages and different root causes."
+          },
+          {
+            "left": "issue:43550",
+            "right": "issue:44190",
+            "accept": false,
+            "reason": "Different subsystems entirely: torch.compile/SDPA model failure vs. local dataset loading in an example script."
+          },
+          {
+            "left": "issue:43644",
+            "right": "issue:44913",
+            "accept": false,
+            "reason": "Non-persistent buffer initialization junk is unrelated to GPTNeoX config reload behavior."
+          },
+          {
+            "left": "issue:44077",
+            "right": "issue:44877",
+            "accept": false,
+            "reason": "Both touch configs, but one is a model-specific post_init policy issue and the other is strict config loading for granite_speech."
+          },
+          {
+            "left": "issue:44496",
+            "right": "issue:45803",
+            "accept": false,
+            "reason": "Unrecognized-model/config metadata error is unrelated to Fuyu's broad exception handling bug."
+          },
+          {
+            "left": "issue:43540",
+            "right": "issue:44734",
+            "accept": false,
+            "reason": "Different multimodal/video-serving failures with different code paths and symptoms."
+          },
+          {
+            "left": "issue:45276",
+            "right": "issue:45584",
+            "accept": false,
+            "reason": "Token-embedding resize propagation and Whisper empty-transcription handling are unrelated."
+          },
+          {
+            "left": "issue:44112",
+            "right": "issue:45600",
+            "accept": false,
+            "reason": "Both affect CI, but they are distinct failures in different files and due to different causes."
+          },
+          {
+            "left": "issue:43650",
+            "right": "issue:44220",
+            "accept": false,
+            "reason": "No meaningful overlap: a placeholder issue vs. an fbank feature extraction bug."
+          },
+          {
+            "left": "issue:43335",
+            "right": "issue:44190",
+            "accept": false,
+            "reason": "Different bug classes: SwitchTransformers config construction vs. image-classification example dataset loading."
+          },
+          {
+            "left": "issue:43116",
+            "right": "issue:44190",
+            "accept": false,
+            "reason": "Both are example-script problems, but one is multi-label classification output logic and the other is local dataset loading."
+          },
+          {
+            "left": "issue:43295",
+            "right": "issue:45276",
+            "accept": false,
+            "reason": "Custom processor/tokenizer regression is unrelated to token embedding resize behavior."
+          },
+          {
+            "left": "issue:45245",
+            "right": "issue:45584",
+            "accept": false,
+            "reason": "Category-count limit error is unrelated to Whisper post-processing on empty transcription."
+          },
+          {
+            "left": "issue:43479",
+            "right": "issue:45375",
+            "accept": false,
+            "reason": "Both are multimodal config issues, but they concern different models and different missing/default fields."
+          },
+          {
+            "left": "issue:36246",
+            "right": "issue:41628",
+            "accept": false,
+            "reason": "Both are import errors, but for different symbols and different modules; not the same missing-export bug."
+          },
+          {
+            "left": "issue:44464",
+            "right": "issue:45563",
+            "accept": false,
+            "reason": "Compiled-forward chunked generation inconsistency is unrelated to a stale warning about num_return_sequences."
+          },
+          {
+            "left": "issue:44933",
+            "right": "issue:45362",
+            "accept": false,
+            "reason": "A missing import from image_utils is not the same as a Qwen3.5 chat crash."
+          },
+          {
+            "left": "issue:44186",
+            "right": "issue:44734",
+            "accept": false,
+            "reason": "Tokenizer padding/truncation crash and KV-cache continuation tensor indexing are different failures."
+          },
+          {
+            "left": "issue:43404",
+            "right": "issue:45216",
+            "accept": false,
+            "reason": "Both involve model weight handling, but one is lm_head tying in Mistral3 and the other is incorrect checkpoint saving in Qwen3.5."
+          },
+          {
+            "left": "issue:44683",
+            "right": "issue:45446",
+            "accept": false,
+            "reason": "Shared flex_attention area, but one is a compiled-runtime failure on newer torch while the other is a version-check/import gating bug."
+          },
+          {
+            "left": "issue:38175",
+            "right": "issue:43065",
+            "accept": false,
+            "reason": "Siglip2 zero probabilities and Sam3PixelDecoder dummy Conv2d are unrelated model-specific bugs."
+          },
+          {
+            "left": "issue:44734",
+            "right": "issue:45632",
+            "accept": false,
+            "reason": "Serve KV-cache continuation indexing bug is unrelated to trust_remote_code cache path collisions."
+          },
+          {
+            "left": "issue:44561",
+            "right": "issue:45632",
+            "accept": false,
+            "reason": "Both mention trust_remote_code, but one is a removed-helper compatibility break and the other is a cache-path collision."
+          },
+          {
+            "left": "issue:44466",
+            "right": "issue:44913",
+            "accept": false,
+            "reason": "Serialized lm_head/tied-weight inconsistency is unrelated to GPTNeoX rotary_pct reload behavior."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 30,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:36246",
+        "issue:38617",
+        "issue:41950",
+        "issue:42175",
+        "issue:42907",
+        "issue:43295",
+        "issue:43299",
+        "issue:43381",
+        "issue:43388",
+        "issue:43441",
+        "issue:43526",
+        "issue:43531",
+        "issue:43643",
+        "issue:43673",
+        "issue:43688",
+        "issue:43827",
+        "issue:43854",
+        "issue:43901",
+        "issue:43976",
+        "issue:44016",
+        "issue:44062",
+        "issue:44075",
+        "issue:44112",
+        "issue:44220",
+        "issue:44242",
+        "issue:44246",
+        "issue:44279",
+        "issue:44336",
+        "issue:44351",
+        "issue:44368",
+        "issue:44496",
+        "issue:44964",
+        "issue:45003",
+        "issue:45020",
+        "issue:45042",
+        "issue:45200",
+        "issue:45216",
+        "issue:45278",
+        "issue:45335",
+        "issue:45357",
+        "issue:45405",
+        "issue:45406",
+        "issue:45563",
+        "issue:45584",
+        "issue:45593",
+        "issue:45600",
+        "issue:45632",
+        "issue:45759"
+      ],
+      "soft_pairs": [
+        "issue:44246|issue:45003",
+        "issue:36246|issue:38617",
+        "issue:43643|issue:44496",
+        "issue:45563|issue:45584",
+        "issue:45406|issue:45632",
+        "issue:41950|issue:44016",
+        "issue:44964|issue:45600",
+        "issue:43388|issue:43688",
+        "issue:42907|issue:43441",
+        "issue:44112|issue:44964",
+        "issue:43295|issue:45335",
+        "issue:43854|issue:44964",
+        "issue:43673|issue:44336",
+        "issue:44016|issue:44220",
+        "issue:44496|issue:45042",
+        "issue:43299|issue:45216",
+        "issue:43531|issue:45600",
+        "issue:43526|issue:43901",
+        "issue:42175|issue:44279",
+        "issue:45335|issue:45584",
+        "issue:44075|issue:44368",
+        "issue:43381|issue:45593",
+        "issue:43299|issue:45357",
+        "issue:43901|issue:44496",
+        "issue:44496|issue:45759",
+        "issue:44351|issue:45278",
+        "issue:44242|issue:45200",
+        "issue:44016|issue:44062",
+        "issue:43388|issue:43827",
+        "issue:43976|issue:45405",
+        "issue:41950|issue:45020"
+      ],
+      "prepared_review_unit_hash": "f27d8f975a3b0c0b2e4bf7f1b34a4ce60d05a41780c656dfa62a905080b0527d",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 31,
+        "serialized_chars": 37050,
+        "estimated_input_tokens": 9263,
+        "estimated_eval_tokens": 18782
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "No duplicate cluster here: the issues span many unrelated regressions, import errors, model-loading problems, docs, and training/runtime bugs. None of the soft pairs look like the same underlying defect.",
+        "confidence": 0.97,
+        "canonical_issue_reason": "No single canonical issue is appropriate; the set is too heterogeneous to treat one report as representative of the rest.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "No strong best issue either; the broadest report is still not a true umbrella for the others, so merging them as duplicates would be misleading.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:44246",
+            "right": "issue:45003",
+            "accept": false,
+            "reason": "One is intermittent import slowness; the other is an unsafe sys.modules access bug. Different symptoms and code paths."
+          },
+          {
+            "left": "issue:36246",
+            "right": "issue:38617",
+            "accept": false,
+            "reason": "Both are ImportErrors, but they are for different missing symbols in different modules with no shared concrete fix."
+          },
+          {
+            "left": "issue:43643",
+            "right": "issue:44496",
+            "accept": false,
+            "reason": "Remote-code config field loss and unrecognized model_type are different loading failures, not the same bug."
+          },
+          {
+            "left": "issue:45563",
+            "right": "issue:45584",
+            "accept": false,
+            "reason": "Stale paging warning and Whisper empty-transcription failure are unrelated generation issues."
+          },
+          {
+            "left": "issue:45406",
+            "right": "issue:45632",
+            "accept": false,
+            "reason": "Processor _tokenizer attribute crash and trust_remote_code cache collision are separate problems."
+          },
+          {
+            "left": "issue:41950",
+            "right": "issue:44016",
+            "accept": false,
+            "reason": "Video pipeline processor lookup bug and a notebook syntax error are unrelated."
+          },
+          {
+            "left": "issue:44964",
+            "right": "issue:45600",
+            "accept": false,
+            "reason": "Phi-4 multimodal loading and stale auto_mappings references to removed configs are different breakages."
+          },
+          {
+            "left": "issue:43388",
+            "right": "issue:43688",
+            "accept": false,
+            "reason": "Batch label dropping in gather_for_metrics and auxiliary-loss normalization are unrelated training/eval bugs."
+          },
+          {
+            "left": "issue:42907",
+            "right": "issue:43441",
+            "accept": false,
+            "reason": "Saving dequantized models and FlashAttention loading failures in Ministral are distinct code paths."
+          },
+          {
+            "left": "issue:44112",
+            "right": "issue:44964",
+            "accept": false,
+            "reason": "A stale GraniteSpeech CI test is not the same as a Phi-4 multimodal model-loading failure."
+          },
+          {
+            "left": "issue:43295",
+            "right": "issue:45335",
+            "accept": false,
+            "reason": "Processor/tokenizer regression and T5Gemma embedding resize behavior are different model API bugs."
+          },
+          {
+            "left": "issue:43854",
+            "right": "issue:44964",
+            "accept": false,
+            "reason": "GLM-4.7-Flash unit-test loading and Phi-4 multimodal loading are separate model-specific failures."
+          },
+          {
+            "left": "issue:43673",
+            "right": "issue:44336",
+            "accept": false,
+            "reason": "Generation cache missing and ANSI output formatting are unrelated subsystems."
+          },
+          {
+            "left": "issue:44016",
+            "right": "issue:44220",
+            "accept": false,
+            "reason": "Notebook syntax error and _torch_extract_fbank_features failure do not share a concrete defect."
+          },
+          {
+            "left": "issue:44496",
+            "right": "issue:45042",
+            "accept": false,
+            "reason": "Unrecognized model config and PIL backend requiring torchvision are different loading/runtime issues."
+          },
+          {
+            "left": "issue:43299",
+            "right": "issue:45216",
+            "accept": false,
+            "reason": "Qwen3VL MoE model loading and Qwen3.5 save_pretrained checkpoint regression are separate bugs."
+          },
+          {
+            "left": "issue:43531",
+            "right": "issue:45600",
+            "accept": false,
+            "reason": "Qwen3-MoE sliding_window behavior and removed Sam3LiteText config references are unrelated."
+          },
+          {
+            "left": "issue:43526",
+            "right": "issue:43901",
+            "accept": false,
+            "reason": "BeitImageProcessorFast label reduction and text-classification docs mismatch are not the same issue."
+          },
+          {
+            "left": "issue:42175",
+            "right": "issue:44279",
+            "accept": false,
+            "reason": "A missing TensorFlow backend dependency and a generic dependency issue are too broad and not the same concrete failure."
+          },
+          {
+            "left": "issue:45335",
+            "right": "issue:45584",
+            "accept": false,
+            "reason": "Embedding resize behavior and Whisper empty transcription generation failure are unrelated."
+          },
+          {
+            "left": "issue:44075",
+            "right": "issue:44368",
+            "accept": false,
+            "reason": "Optimizer SGD args ignored and a tie_word_embeddings warning are different config/runtime concerns."
+          },
+          {
+            "left": "issue:43381",
+            "right": "issue:45593",
+            "accept": false,
+            "reason": "Gradient checkpointing in eval mode and D-FINE auxiliary-loss omission are unrelated behaviors."
+          },
+          {
+            "left": "issue:43299",
+            "right": "issue:45357",
+            "accept": false,
+            "reason": "These are both Qwen3.5 regressions, but one is model loading and the other is save_pretrained key serialization; not one bug."
+          },
+          {
+            "left": "issue:43901",
+            "right": "issue:44496",
+            "accept": false,
+            "reason": "Docs mentioning return_all_scores and an unrecognized model config error do not share a fix."
+          },
+          {
+            "left": "issue:44351",
+            "right": "issue:45278",
+            "accept": false,
+            "reason": "A specific HybridCache import error is not the same concrete bug as a broad report of many import errors."
+          },
+          {
+            "left": "issue:44242",
+            "right": "issue:45200",
+            "accept": false,
+            "reason": "Dropped label elements in metrics gathering and Gemma 4 mm_token_type_ids defaults are unrelated."
+          },
+          {
+            "left": "issue:44016",
+            "right": "issue:44062",
+            "accept": false,
+            "reason": "Notebook syntax error and AddedToken keyword conflict are unrelated."
+          },
+          {
+            "left": "issue:43388",
+            "right": "issue:43827",
+            "accept": false,
+            "reason": "Metrics label handling and docs still referencing pipeline() after removals are different issues."
+          },
+          {
+            "left": "issue:43976",
+            "right": "issue:45405",
+            "accept": false,
+            "reason": "Python version incompatibility and an unreleased PEFT minimum version bump are unrelated."
+          },
+          {
+            "left": "issue:41950",
+            "right": "issue:45020",
+            "accept": false,
+            "reason": "Video-classification pipeline image-processor lookup and remote_code breakage are different bugs."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 31,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:32090",
+        "issue:35141",
+        "issue:38175",
+        "issue:39692",
+        "issue:41950",
+        "issue:42994",
+        "issue:43065",
+        "issue:43295",
+        "issue:43299",
+        "issue:43317",
+        "issue:43441",
+        "issue:43450",
+        "issue:43479",
+        "issue:43525",
+        "issue:43577",
+        "issue:43650",
+        "issue:43723",
+        "issue:43824",
+        "issue:43854",
+        "issue:43901",
+        "issue:43976",
+        "issue:44162",
+        "issue:44261",
+        "issue:44263",
+        "issue:44279",
+        "issue:44297",
+        "issue:44462",
+        "issue:44466",
+        "issue:44521",
+        "issue:44617",
+        "issue:44623",
+        "issue:44683",
+        "issue:44779",
+        "issue:44898",
+        "issue:44933",
+        "issue:44987",
+        "issue:45081",
+        "issue:45216",
+        "issue:45276",
+        "issue:45290",
+        "issue:45325",
+        "issue:45397",
+        "issue:45405",
+        "issue:45542",
+        "issue:45600",
+        "issue:45712",
+        "issue:45823",
+        "issue:45834"
+      ],
+      "soft_pairs": [
+        "issue:43824|issue:44933",
+        "issue:43577|issue:45216",
+        "issue:43479|issue:43525",
+        "issue:41950|issue:45823",
+        "issue:43299|issue:43854",
+        "issue:44683|issue:45542",
+        "issue:44279|issue:45405",
+        "issue:39692|issue:43901",
+        "issue:38175|issue:44617",
+        "issue:32090|issue:35141",
+        "issue:44462|issue:45081",
+        "issue:43295|issue:44779",
+        "issue:44297|issue:44462",
+        "issue:41950|issue:43650",
+        "issue:43976|issue:44279",
+        "issue:43976|issue:44987",
+        "issue:45600|issue:45712",
+        "issue:44462|issue:44779",
+        "issue:44521|issue:45325",
+        "issue:43441|issue:45834",
+        "issue:43450|issue:44279",
+        "issue:44261|issue:44263",
+        "issue:42994|issue:43317",
+        "issue:43065|issue:44898",
+        "issue:44466|issue:45276",
+        "issue:43650|issue:45290",
+        "issue:43441|issue:44683",
+        "issue:43723|issue:44162",
+        "issue:44623|issue:45397"
+      ],
+      "prepared_review_unit_hash": "2bf4344e4347861c0358e0a8d2d62c372cce169323e8c84f1db7d0aa059c1845",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 29,
+        "serialized_chars": 36327,
+        "estimated_input_tokens": 9082,
+        "estimated_eval_tokens": 18420
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "These items are a mixed bag of mostly unrelated Transformers bugs across tokenizers, multimodal processing, loading/saving, and backend failures. None of the soft-similarity pairs looks like the same concrete bug or fix, so I reject all soft edges. Best canonical issue is the clearest, most self-contained tokenizer-loading regression.",
+        "confidence": 0.84,
+        "canonical_issue_reason": "#44462 is a well-scoped, reproducible core API bug with a clear failure mode in AutoTokenizer loading from the repo.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "#44462 is the strongest representative issue here because it isolates one concrete code path and is better specified than the more model-specific or documentation-oriented reports.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:43824",
+            "right": "issue:44933",
+            "accept": false,
+            "reason": "Different import failures in unrelated modules: Qwen2.5-VL class export vs a missing image_utils symbol."
+          },
+          {
+            "left": "issue:43577",
+            "right": "issue:45216",
+            "accept": false,
+            "reason": "Both involve loading behavior, but one is dtype propagation in BLIP-2 and the other is Qwen3.5 checkpoint saving corruption."
+          },
+          {
+            "left": "issue:43479",
+            "right": "issue:43525",
+            "accept": false,
+            "reason": "Different model config bugs: default multimodal sub-config initialization vs missing pad_token_id on Llama4Config."
+          },
+          {
+            "left": "issue:41950",
+            "right": "issue:45823",
+            "accept": false,
+            "reason": "Unrelated multimodal areas: video-classification processor lookup vs Gemma4 device_map forward mismatch."
+          },
+          {
+            "left": "issue:43299",
+            "right": "issue:43854",
+            "accept": false,
+            "reason": "Different model-loading regressions affecting different architectures and code paths."
+          },
+          {
+            "left": "issue:44683",
+            "right": "issue:45542",
+            "accept": false,
+            "reason": "Completely different failures: compiled flex_attention on torch 2.9+ vs tf backend detection without TensorFlow."
+          },
+          {
+            "left": "issue:44279",
+            "right": "issue:45405",
+            "accept": false,
+            "reason": "General dependency complaint vs a specific PEFT version bump; not the same bug."
+          },
+          {
+            "left": "issue:39692",
+            "right": "issue:43901",
+            "accept": false,
+            "reason": "Docs example/model-processor mismatch and quantization errors are unrelated to a stale pipeline docs parameter mention."
+          },
+          {
+            "left": "issue:38175",
+            "right": "issue:44617",
+            "accept": false,
+            "reason": "Different symptoms and subsystems: zero probabilities in SigLIP2 vs Sam3Video OOM."
+          },
+          {
+            "left": "issue:32090",
+            "right": "issue:35141",
+            "accept": false,
+            "reason": "Trainer broadcast TypeError is unrelated to embedding resize causing post_init reinitialization."
+          },
+          {
+            "left": "issue:44462",
+            "right": "issue:45081",
+            "accept": false,
+            "reason": "Both touch tokenizers, but one is repo tokenizer discovery and the other is a Mistral regex patch crash."
+          },
+          {
+            "left": "issue:43295",
+            "right": "issue:44779",
+            "accept": false,
+            "reason": "Different v5 tokenizer regressions with different APIs and models; not one concrete underlying bug."
+          },
+          {
+            "left": "issue:44297",
+            "right": "issue:44462",
+            "accept": false,
+            "reason": "Tokenizer class metadata on save_pretrained is distinct from AutoTokenizer ignoring tokenizer.json on load."
+          },
+          {
+            "left": "issue:41950",
+            "right": "issue:43650",
+            "accept": false,
+            "reason": "Video-classification processor bug is unrelated to the placeholder 'ADD THE DATA' issue."
+          },
+          {
+            "left": "issue:43976",
+            "right": "issue:44279",
+            "accept": false,
+            "reason": "Python version compatibility failure is not the same as a generic dependency issue."
+          },
+          {
+            "left": "issue:43976",
+            "right": "issue:44987",
+            "accept": false,
+            "reason": "Python-version support and model-load failure are different problems."
+          },
+          {
+            "left": "issue:45600",
+            "right": "issue:45712",
+            "accept": false,
+            "reason": "Both are CI cleanup issues, but they target different files and different defects."
+          },
+          {
+            "left": "issue:44462",
+            "right": "issue:44779",
+            "accept": false,
+            "reason": "Both tokenizer-related, but one is loading repo tokenizer assets and the other is incorrect Deepseek tokenization logic."
+          },
+          {
+            "left": "issue:44521",
+            "right": "issue:45325",
+            "accept": false,
+            "reason": "Multimodal chat masks and Qwen2.5-VL rope indexing are separate bugs in different code paths."
+          },
+          {
+            "left": "issue:43441",
+            "right": "issue:45834",
+            "accept": false,
+            "reason": "Ministral FlashAttention failure and Kosmos2.5 OCR index error are unrelated."
+          },
+          {
+            "left": "issue:43450",
+            "right": "issue:44279",
+            "accept": false,
+            "reason": "Video processor shape bug is unrelated to the dependency complaint."
+          },
+          {
+            "left": "issue:44261",
+            "right": "issue:44263",
+            "accept": false,
+            "reason": "Different model internals: MLA layernorm epsilon config vs torch.split return handling in GlmMoeDsaIndexer."
+          },
+          {
+            "left": "issue:42994",
+            "right": "issue:43317",
+            "accept": false,
+            "reason": "Quantized model saving and dequantized model offload loading are different lifecycle failures."
+          },
+          {
+            "left": "issue:43065",
+            "right": "issue:44898",
+            "accept": false,
+            "reason": "Dummy Conv2d in Sam3PixelDecoder is unrelated to Perceiver non-default-resolution failure."
+          },
+          {
+            "left": "issue:44466",
+            "right": "issue:45276",
+            "accept": false,
+            "reason": "Both mention weights, but one is device-dependent serialization and the other is resize_token_embeddings not updating Gemma4 outputs."
+          },
+          {
+            "left": "issue:43650",
+            "right": "issue:45290",
+            "accept": false,
+            "reason": "Placeholder issue has no relation to chat template crashes with empty tool-call assistant messages."
+          },
+          {
+            "left": "issue:43441",
+            "right": "issue:44683",
+            "accept": false,
+            "reason": "FlashAttention-specific model failure is not the same as a torch-version compiled flex_attention failure."
+          },
+          {
+            "left": "issue:43723",
+            "right": "issue:44162",
+            "accept": false,
+            "reason": "Tokenizer loading in v5 and ESM2 model breakage are separate regressions."
+          },
+          {
+            "left": "issue:44623",
+            "right": "issue:45397",
+            "accept": false,
+            "reason": "Missing processor save files and Gemma4 zero3 from_pretrained are different save/load bugs."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 32,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:16998",
+        "issue:22355",
+        "issue:41628",
+        "issue:41950",
+        "issue:42491",
+        "issue:43122",
+        "issue:43262",
+        "issue:43299",
+        "issue:43352",
+        "issue:43450",
+        "issue:43525",
+        "issue:43643",
+        "issue:43650",
+        "issue:43653",
+        "issue:43673",
+        "issue:43761",
+        "issue:43784",
+        "issue:43881",
+        "issue:43901",
+        "issue:43976",
+        "issue:43994",
+        "issue:44016",
+        "issue:44188",
+        "issue:44220",
+        "issue:44242",
+        "issue:44246",
+        "issue:44291",
+        "issue:44295",
+        "issue:44297",
+        "issue:44336",
+        "issue:44361",
+        "issue:44373",
+        "issue:44496",
+        "issue:44521",
+        "issue:44589",
+        "issue:44617",
+        "issue:44623",
+        "issue:44671",
+        "issue:44749",
+        "issue:44779",
+        "issue:44987",
+        "issue:45250",
+        "issue:45362",
+        "issue:45397",
+        "issue:45542",
+        "issue:45593",
+        "issue:45823"
+      ],
+      "soft_pairs": [
+        "issue:44623|issue:44749",
+        "issue:43653|issue:45593",
+        "issue:43784|issue:45542",
+        "issue:44671|issue:44779",
+        "issue:44016|issue:45823",
+        "issue:44361|issue:45593",
+        "issue:44987|issue:45542",
+        "issue:43901|issue:43994",
+        "issue:43122|issue:43650",
+        "issue:44016|issue:44291",
+        "issue:43262|issue:44521",
+        "issue:44188|issue:44336",
+        "issue:43673|issue:43761",
+        "issue:43525|issue:43881",
+        "issue:44297|issue:45397",
+        "issue:43525|issue:43643",
+        "issue:43450|issue:44617",
+        "issue:42491|issue:43299",
+        "issue:44589|issue:45362",
+        "issue:41628|issue:43976",
+        "issue:43901|issue:44246",
+        "issue:44336|issue:44373",
+        "issue:44242|issue:45593",
+        "issue:43352|issue:45250",
+        "issue:16998|issue:22355",
+        "issue:44295|issue:45397",
+        "issue:44246|issue:44496",
+        "issue:41950|issue:44220"
+      ],
+      "prepared_review_unit_hash": "fa53bef158986be33eb12bbf2e87691f694b80c6a51df29d435538d71f03dafa",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 47,
+        "item_count": 47,
+        "soft_pair_count": 28,
+        "serialized_chars": 35147,
+        "estimated_input_tokens": 8787,
+        "estimated_eval_tokens": 17830
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "This cluster is mostly noise: the soft-linked issues span unrelated subsystems (tokenizers, model loading, processors, docs, and performance), so almost none look like true duplicates.",
+        "confidence": 0.82,
+        "canonical_issue_reason": "issue:44521 is the best representative because it is the most active open issue in the processor/chat_template area and has the strongest inbound interest among the set.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "issue:44521 is the most suitable global issue target for this cluster, but only as a representative theme, not because the linked items are actually the same bug.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:44623",
+            "right": "issue:44749",
+            "accept": false,
+            "reason": "One is a processor save_pretrained file-generation bug; the other is a data-filtering slowdown regression. Different symptoms and code paths."
+          },
+          {
+            "left": "issue:43653",
+            "right": "issue:45593",
+            "accept": false,
+            "reason": "BigBirdTokenizer special-token decode corruption is unrelated to D-FINE auxiliary-loss handling."
+          },
+          {
+            "left": "issue:43784",
+            "right": "issue:45542",
+            "accept": false,
+            "reason": "Import-time NameError in sentence-transformers integration does not match a TensorFlow/tensorboard backend selection bug."
+          },
+          {
+            "left": "issue:44671",
+            "right": "issue:44779",
+            "accept": false,
+            "reason": "Both affect outputs, but one is CamemBERT MLM prediction quality and the other is Deepseek tokenization correctness; different models and root causes."
+          },
+          {
+            "left": "issue:44016",
+            "right": "issue:45823",
+            "accept": false,
+            "reason": "Notebook syntax error is unrelated to a Gemma4 device-mismatch forward failure."
+          },
+          {
+            "left": "issue:44361",
+            "right": "issue:45593",
+            "accept": false,
+            "reason": "MLukeTokenizer task AttributeError and D-FINE auxiliary-loss omission are unrelated."
+          },
+          {
+            "left": "issue:44987",
+            "right": "issue:45542",
+            "accept": false,
+            "reason": "Model loading regression on transformers>=5.1.0 is not the same as a missing/undefined tf backend error."
+          },
+          {
+            "left": "issue:43901",
+            "right": "issue:43994",
+            "accept": false,
+            "reason": "Documentation mismatch about return_all_scores is not the same as an incorrect model/pipeline result bug."
+          },
+          {
+            "left": "issue:43122",
+            "right": "issue:43650",
+            "accept": false,
+            "reason": "Tokenization behavior change and a vague \"ADD THE DATA\" report do not describe the same concrete defect."
+          },
+          {
+            "left": "issue:44016",
+            "right": "issue:44291",
+            "accept": false,
+            "reason": "Notebook syntax error and init_empty_weights argument handling are unrelated."
+          },
+          {
+            "left": "issue:43262",
+            "right": "issue:44521",
+            "accept": false,
+            "reason": "Both touch apply_chat_template, but one is an audio sampling-rate default and the other is multimodal assistant_masks; same API surface, different bug."
+          },
+          {
+            "left": "issue:44188",
+            "right": "issue:44336",
+            "accept": false,
+            "reason": "torch.compile attention-kernel divergence is unrelated to ANSI codes emitted by loading_report."
+          },
+          {
+            "left": "issue:43673",
+            "right": "issue:43761",
+            "accept": false,
+            "reason": "Generation cache missing in chunked_prefill and CLIPVision hidden_states handling are different model-path regressions."
+          },
+          {
+            "left": "issue:43525",
+            "right": "issue:43881",
+            "accept": false,
+            "reason": "Missing pad_token_id on Llama4Config and a glm-4v-9b loading failure are separate model-specific issues."
+          },
+          {
+            "left": "issue:44297",
+            "right": "issue:45397",
+            "accept": false,
+            "reason": "Tokenizer save_pretrained metadata mismatch is unrelated to Gemma-4 zero3 loading."
+          },
+          {
+            "left": "issue:43525",
+            "right": "issue:43643",
+            "accept": false,
+            "reason": "A missing config attribute and trust_remote_code dropping fields are different config-loading bugs."
+          },
+          {
+            "left": "issue:43450",
+            "right": "issue:44617",
+            "accept": false,
+            "reason": "Incorrect batched video-processor shape and CUDA OOM are not the same failure."
+          },
+          {
+            "left": "issue:42491",
+            "right": "issue:43299",
+            "accept": false,
+            "reason": "Both mention Qwen/MoE on v5, but one is LoRA reuse incompatibility and the other is model loading failure; too different to merge."
+          },
+          {
+            "left": "issue:44589",
+            "right": "issue:45362",
+            "accept": false,
+            "reason": "Float8 storage lookup failure is unrelated to Qwen3.5 chat crashes."
+          },
+          {
+            "left": "issue:41628",
+            "right": "issue:43976",
+            "accept": false,
+            "reason": "Missing AutoImageProcessor import and a Python-version compatibility regression are different issues."
+          },
+          {
+            "left": "issue:43901",
+            "right": "issue:44246",
+            "accept": false,
+            "reason": "A docs mismatch about return_all_scores is unrelated to intermittent import slowness."
+          },
+          {
+            "left": "issue:44336",
+            "right": "issue:44373",
+            "accept": false,
+            "reason": "ANSI-color output in loading_report and a wrong docstring for position_ids are separate documentation/UX issues."
+          },
+          {
+            "left": "issue:44242",
+            "right": "issue:45593",
+            "accept": false,
+            "reason": "Both involve losses, but one is MoE load-balancing loss gating and the other is D-FINE auxiliary-loss omission."
+          },
+          {
+            "left": "issue:43352",
+            "right": "issue:45250",
+            "accept": false,
+            "reason": "A lack of Flash Attention 2 support and a generic Flash Attention issue are too broad to be the same concrete bug."
+          },
+          {
+            "left": "issue:16998",
+            "right": "issue:22355",
+            "accept": false,
+            "reason": "model_max_length confusion on DeBERTa-V3 is unrelated to the missing transformers.onnx module."
+          },
+          {
+            "left": "issue:44295",
+            "right": "issue:45397",
+            "accept": false,
+            "reason": "position_ids buffer reading error and Gemma-4 zero3 loading are different failures."
+          },
+          {
+            "left": "issue:44246",
+            "right": "issue:44496",
+            "accept": false,
+            "reason": "Import slowdown and an unrecognized-model/config model_type error are unrelated."
+          },
+          {
+            "left": "issue:41950",
+            "right": "issue:44220",
+            "accept": false,
+            "reason": "Video-classification image-processor lookup and _torch_extract_fbank_features() are different subsystems (video vs audio)."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 33,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:36246",
+        "issue:42757",
+        "issue:42886",
+        "issue:43065",
+        "issue:43232",
+        "issue:43257",
+        "issue:43262",
+        "issue:43408",
+        "issue:43475",
+        "issue:43526",
+        "issue:43540",
+        "issue:43596",
+        "issue:43673",
+        "issue:43824",
+        "issue:43901",
+        "issue:43976",
+        "issue:44162",
+        "issue:44242",
+        "issue:44246",
+        "issue:44295",
+        "issue:44336",
+        "issue:44351",
+        "issue:44485",
+        "issue:44496",
+        "issue:44561",
+        "issue:44749",
+        "issue:44779",
+        "issue:44857",
+        "issue:44871",
+        "issue:44877",
+        "issue:44964",
+        "issue:44987",
+        "issue:44995",
+        "issue:45003",
+        "issue:45137",
+        "issue:45200",
+        "issue:45229",
+        "issue:45292",
+        "issue:45362",
+        "issue:45375",
+        "issue:45459",
+        "issue:45561",
+        "issue:45563",
+        "issue:45584",
+        "issue:45588",
+        "issue:45701",
+        "issue:45759"
+      ],
+      "soft_pairs": [
+        "issue:36246|issue:42757",
+        "issue:36246|issue:44351",
+        "issue:44336|issue:44485",
+        "issue:43673|issue:44561",
+        "issue:43824|issue:43976",
+        "issue:43976|issue:44964",
+        "issue:44242|issue:44857",
+        "issue:44246|issue:44871",
+        "issue:45362|issue:45375",
+        "issue:44162|issue:45200",
+        "issue:44295|issue:44749",
+        "issue:44779|issue:45292",
+        "issue:44877|issue:45375",
+        "issue:45292|issue:45584",
+        "issue:45292|issue:45701",
+        "issue:45561|issue:45563",
+        "issue:43901|issue:45003",
+        "issue:44987|issue:45459",
+        "issue:36246|issue:43596",
+        "issue:43232|issue:44336",
+        "issue:43526|issue:45003",
+        "issue:45375|issue:45759",
+        "issue:44496|issue:45003",
+        "issue:43065|issue:43408",
+        "issue:43257|issue:45137",
+        "issue:44995|issue:45200",
+        "issue:43262|issue:44246",
+        "issue:43475|issue:43540",
+        "issue:42886|issue:45229",
+        "issue:44995|issue:45588"
+      ],
+      "prepared_review_unit_hash": "2111255e8cfdd61ea0f1eeb367652ca0b3b8c070da1bb66989f5f72fc79e4a6a",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 47,
+        "item_count": 47,
+        "soft_pair_count": 30,
+        "serialized_chars": 36095,
+        "estimated_input_tokens": 9024,
+        "estimated_eval_tokens": 18304
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "This cluster is mostly a mix of unrelated Transformers bugs that share superficial similarity around imports, configs, generation, and model-loading regressions. No soft pair looks like a true duplicate fix path, so I rejected all edges. Among the issues, the clearest representative is the Qwen2.5-VL import failure because it has the strongest cross-reference signal and a concrete user-facing regression.",
+        "confidence": 0.86,
+        "canonical_issue_reason": "issue:43824 is the strongest representative of the cluster: it is a concrete, user-visible import regression, has the most inbound references, and the title/body pattern is central to several nearby import-failure reports.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "issue:43824 is the best overall issue to anchor the cluster because it is the most broadly referenced and most actionable concrete bug in the set.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:36246",
+            "right": "issue:42757",
+            "accept": false,
+            "reason": "Different missing symbols from different packages/modules; same broad symptom, but not the same underlying bug."
+          },
+          {
+            "left": "issue:36246",
+            "right": "issue:44351",
+            "accept": false,
+            "reason": "Both are import errors, but for unrelated classes/symbols and different failure paths."
+          },
+          {
+            "left": "issue:44336",
+            "right": "issue:44485",
+            "accept": false,
+            "reason": "ANSI/color output bug in loading_report vs a GLM-5 RoPE modeling discussion; unrelated."
+          },
+          {
+            "left": "issue:43673",
+            "right": "issue:44561",
+            "accept": false,
+            "reason": "Both are v5 regressions, but one is generation cache/chunked_prefill and the other is a removed utility breaking trust_remote_code; different code paths."
+          },
+          {
+            "left": "issue:43824",
+            "right": "issue:43976",
+            "accept": false,
+            "reason": "Specific Qwen2.5-VL import failure vs a Python/version compatibility problem; not the same bug."
+          },
+          {
+            "left": "issue:43976",
+            "right": "issue:44964",
+            "accept": false,
+            "reason": "Both mention loading/version issues, but one is Python support and the other is a multimodal model load failure."
+          },
+          {
+            "left": "issue:44242",
+            "right": "issue:44857",
+            "accept": false,
+            "reason": "Different model/loss bugs: router-loss gating vs AMP/CUDA crash in LwDetrImageLoss."
+          },
+          {
+            "left": "issue:44246",
+            "right": "issue:44871",
+            "accept": false,
+            "reason": "Import latency bug vs Gemma EOS-token config inconsistency; unrelated."
+          },
+          {
+            "left": "issue:45362",
+            "right": "issue:45375",
+            "accept": false,
+            "reason": "Same Qwen family, but different concrete problems: chat crash vs a missing config field being dropped."
+          },
+          {
+            "left": "issue:44162",
+            "right": "issue:45200",
+            "accept": false,
+            "reason": "ESM2 breakage and Gemma4 mm_token_type_ids behavior are distinct model-specific issues."
+          },
+          {
+            "left": "issue:44295",
+            "right": "issue:44749",
+            "accept": false,
+            "reason": "One is a position_ids buffer access error; the other is a tokenizer/performance regression across versions."
+          },
+          {
+            "left": "issue:44779",
+            "right": "issue:45292",
+            "accept": false,
+            "reason": "Tokenizer output regression and resize_token_embeddings not updating output embeddings are different bugs."
+          },
+          {
+            "left": "issue:44877",
+            "right": "issue:45375",
+            "accept": false,
+            "reason": "Both involve config strictness, but they affect different model configs and different failure modes."
+          },
+          {
+            "left": "issue:45292",
+            "right": "issue:45584",
+            "accept": false,
+            "reason": "Embedding resize bug vs Whisper empty-transcription generation failure; not the same change."
+          },
+          {
+            "left": "issue:45292",
+            "right": "issue:45701",
+            "accept": false,
+            "reason": "Resize-token-embeddings behavior and tokenization changes are separate concerns."
+          },
+          {
+            "left": "issue:45561",
+            "right": "issue:45563",
+            "accept": false,
+            "reason": "Test-worker file race vs a stale generation warning; unrelated."
+          },
+          {
+            "left": "issue:43901",
+            "right": "issue:45003",
+            "accept": false,
+            "reason": "Docs mismatch for a pipeline vs unsafe sys.modules access in modeling_utils; different issues entirely."
+          },
+          {
+            "left": "issue:44987",
+            "right": "issue:45459",
+            "accept": false,
+            "reason": "Model load failure on transformers>=5.1.0 vs tokenizer error masking when protobuf is absent; different root causes."
+          },
+          {
+            "left": "issue:36246",
+            "right": "issue:43596",
+            "accept": false,
+            "reason": "Qwen import failure vs DeepSpeed ZeRO3/BertModel initialization crash; not the same bug."
+          },
+          {
+            "left": "issue:43232",
+            "right": "issue:44336",
+            "accept": false,
+            "reason": "Generation kwargs/update bug vs ANSI escape output problem; unrelated."
+          },
+          {
+            "left": "issue:43526",
+            "right": "issue:45003",
+            "accept": false,
+            "reason": "BeitImageProcessorFast label reduction bug vs sys.modules import handling; different code paths."
+          },
+          {
+            "left": "issue:45375",
+            "right": "issue:45759",
+            "accept": false,
+            "reason": "Both concern Qwen multimodal config handling, but one is a missing field dropped by strict config parsing and the other is text_config unwrapping in from_config; not the same fix."
+          },
+          {
+            "left": "issue:44496",
+            "right": "issue:45003",
+            "accept": false,
+            "reason": "Unrecognized model/config.json problem vs unsafe sys.modules access; unrelated."
+          },
+          {
+            "left": "issue:43065",
+            "right": "issue:43408",
+            "accept": false,
+            "reason": "Same SAM3 area, but one is a dummy Conv2d design choice and the other is a model-type warning; not the same underlying bug."
+          },
+          {
+            "left": "issue:43257",
+            "right": "issue:45137",
+            "accept": false,
+            "reason": "Different DeepSpeed-related failures: Qwen3 MOE weight conversion vs ZeRO3 deque underflow."
+          },
+          {
+            "left": "issue:44995",
+            "right": "issue:45200",
+            "accept": false,
+            "reason": "GlmMoeDsa stale cache crash and Gemma4 token-type defaults are unrelated."
+          },
+          {
+            "left": "issue:43262",
+            "right": "issue:44246",
+            "accept": false,
+            "reason": "Audio chat-template sampling-rate default bug vs import-time performance issue; unrelated."
+          },
+          {
+            "left": "issue:43475",
+            "right": "issue:43540",
+            "accept": false,
+            "reason": "Different multimodal model failures: missing vision encoder output attribute vs Qwen3OmniMoe video-input ValueError."
+          },
+          {
+            "left": "issue:42886",
+            "right": "issue:45229",
+            "accept": false,
+            "reason": "Offline cache/tokenizer loading bug vs multi-GPU CUDA OOM; not the same issue."
+          },
+          {
+            "left": "issue:44995",
+            "right": "issue:45588",
+            "accept": false,
+            "reason": "Stale indexer cache crash in GlmMoeDsa vs flash_attention AttributeError for sink-less models; different bugs."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 34,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:28282",
+        "issue:30990",
+        "issue:41628",
+        "issue:42371",
+        "issue:42994",
+        "issue:43232",
+        "issue:43452",
+        "issue:43475",
+        "issue:43482",
+        "issue:43525",
+        "issue:43643",
+        "issue:43646",
+        "issue:43653",
+        "issue:43673",
+        "issue:43827",
+        "issue:43931",
+        "issue:44038",
+        "issue:44060",
+        "issue:44077",
+        "issue:44162",
+        "issue:44297",
+        "issue:44361",
+        "issue:44368",
+        "issue:44466",
+        "issue:44623",
+        "issue:44734",
+        "issue:44749",
+        "issue:44869",
+        "issue:44933",
+        "issue:44977",
+        "issue:45003",
+        "issue:45081",
+        "issue:45092",
+        "issue:45103",
+        "issue:45137",
+        "issue:45230",
+        "issue:45310",
+        "issue:45335",
+        "issue:45356",
+        "issue:45375",
+        "issue:45381",
+        "issue:45397",
+        "issue:45459",
+        "issue:45478",
+        "issue:45561",
+        "issue:45563",
+        "issue:45600",
+        "issue:45656"
+      ],
+      "soft_pairs": [
+        "issue:44466|issue:45335",
+        "issue:43653|issue:44162",
+        "issue:44368|issue:45563",
+        "issue:44368|issue:44977",
+        "issue:45397|issue:45656",
+        "issue:44060|issue:44077",
+        "issue:45003|issue:45561",
+        "issue:43673|issue:44734",
+        "issue:45137|issue:45600",
+        "issue:43931|issue:44368",
+        "issue:44933|issue:45459",
+        "issue:44869|issue:45356",
+        "issue:44297|issue:44623",
+        "issue:43646|issue:44162",
+        "issue:44361|issue:44749",
+        "issue:43232|issue:43673",
+        "issue:43673|issue:43827",
+        "issue:43452|issue:45003",
+        "issue:44623|issue:45656",
+        "issue:28282|issue:30990",
+        "issue:42994|issue:43482",
+        "issue:44749|issue:45310",
+        "issue:44749|issue:45656",
+        "issue:42371|issue:45092",
+        "issue:43475|issue:43643",
+        "issue:43475|issue:45381",
+        "issue:45230|issue:45397",
+        "issue:44038|issue:44623",
+        "issue:41628|issue:44933",
+        "issue:45081|issue:45103",
+        "issue:42994|issue:43525",
+        "issue:45375|issue:45478"
+      ],
+      "prepared_review_unit_hash": "74a998a8e668aaddb40e7143cbdd829488f5ffc5e62997087cab130c6f72e06b",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 32,
+        "serialized_chars": 37312,
+        "estimated_input_tokens": 9328,
+        "estimated_eval_tokens": 18912
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "This cluster is mostly heterogeneous false-positive similarity: many issues are in the same project/version band but describe different bugs. The only pair that looks plausibly like the same underlying regression is the Qwen3.5-Moe from_pretrained/config field case.",
+        "confidence": 0.58,
+        "canonical_issue_reason": "issue:45310 is the cleanest representative of the Qwen3.5-Moe from_pretrained regression; it matches the later same-title report and is the most direct anchor for the only plausible duplicate mini-cluster.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "issue:45310 is the best issue candidate because it is an exact, concrete regression report rather than a broad symptom summary, and it best represents the one likely duplicate topic in this set.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:44466",
+            "right": "issue:45335",
+            "accept": false,
+            "reason": "Both touch tied embeddings, but one is serialization/device-dependent while the other is resize_token_embeddings behavior; different code paths."
+          },
+          {
+            "left": "issue:43653",
+            "right": "issue:44162",
+            "accept": false,
+            "reason": "Different models and failures: BigBirdTokenizer special-token registration vs ESM2 breakage."
+          },
+          {
+            "left": "issue:44368",
+            "right": "issue:45563",
+            "accept": false,
+            "reason": "Both are warnings, but one is about tie_word_embeddings and the other is a stale num_return_sequences warning; not the same bug."
+          },
+          {
+            "left": "issue:44368",
+            "right": "issue:44977",
+            "accept": false,
+            "reason": "Different symptoms and subsystems: tied-weights warning vs flash-attention generation failure."
+          },
+          {
+            "left": "issue:45397",
+            "right": "issue:45656",
+            "accept": false,
+            "reason": "Gemma-4 Zero3 loading and DeepSpeed optimizer double-step are unrelated runtime issues."
+          },
+          {
+            "left": "issue:44060",
+            "right": "issue:44077",
+            "accept": false,
+            "reason": "Different model/config validation bugs; no shared failing code path is evident."
+          },
+          {
+            "left": "issue:45003",
+            "right": "issue:45561",
+            "accept": false,
+            "reason": "sys.modules access in modeling_utils and xdist captured_info.txt races are unrelated."
+          },
+          {
+            "left": "issue:43673",
+            "right": "issue:44734",
+            "accept": false,
+            "reason": "Both involve generation/cache, but one is chunked_prefill cache handling and the other is serve KV-cache continuation indexing; not the same bug."
+          },
+          {
+            "left": "issue:45137",
+            "right": "issue:45600",
+            "accept": false,
+            "reason": "DeepSpeed deque underflow and auto_mappings references to removed configs are unrelated."
+          },
+          {
+            "left": "issue:43931",
+            "right": "issue:44368",
+            "accept": false,
+            "reason": "Qwen3-VL weight-shape mismatch is unrelated to the tied-embeddings warning issue."
+          },
+          {
+            "left": "issue:44933",
+            "right": "issue:45459",
+            "accept": false,
+            "reason": "Removed image_utils import and protobuf tokenization error masking are different import-time failures."
+          },
+          {
+            "left": "issue:44869",
+            "right": "issue:45356",
+            "accept": false,
+            "reason": "Both are tokenizer-related, but they affect different models and different decode/codec paths."
+          },
+          {
+            "left": "issue:44297",
+            "right": "issue:44623",
+            "accept": false,
+            "reason": "Both are save_pretrained regressions, but tokenizer metadata mismatch and processor missing files are distinct serialization bugs."
+          },
+          {
+            "left": "issue:43646",
+            "right": "issue:44162",
+            "accept": false,
+            "reason": "Broad v5 model-init regressions, but the reports describe different affected models and failure modes."
+          },
+          {
+            "left": "issue:44361",
+            "right": "issue:44749",
+            "accept": false,
+            "reason": "Tokenizer AttributeError and a performance regression are not the same underlying issue."
+          },
+          {
+            "left": "issue:43232",
+            "right": "issue:43673",
+            "accept": false,
+            "reason": "Both mention generation internals, but sync_gpus kwargs update and chunked_prefill cache missing are different bugs."
+          },
+          {
+            "left": "issue:43673",
+            "right": "issue:43827",
+            "accept": false,
+            "reason": "Generation cache internals vs docs referencing removed pipeline APIs; unrelated."
+          },
+          {
+            "left": "issue:43452",
+            "right": "issue:45003",
+            "accept": false,
+            "reason": "gguf_file loading failures and unsafe sys.modules access are separate code paths."
+          },
+          {
+            "left": "issue:44623",
+            "right": "issue:45656",
+            "accept": false,
+            "reason": "Processor serialization and DeepSpeed optimizer stepping are unrelated."
+          },
+          {
+            "left": "issue:28282",
+            "right": "issue:30990",
+            "accept": false,
+            "reason": "PyTorch-missing ImportError and Sentence Transformers loading hang are different environment/runtime problems."
+          },
+          {
+            "left": "issue:42994",
+            "right": "issue:43482",
+            "accept": false,
+            "reason": "Quantized model saving and GGUF loading failures are related only loosely by format, not the same bug."
+          },
+          {
+            "left": "issue:44749",
+            "right": "issue:45310",
+            "accept": false,
+            "reason": "Performance regression after upgrade vs Qwen3.5-Moe from_pretrained error are different issues."
+          },
+          {
+            "left": "issue:44749",
+            "right": "issue:45656",
+            "accept": false,
+            "reason": "Data filtering slowdown and DeepSpeed optimizer double-step are unrelated."
+          },
+          {
+            "left": "issue:42371",
+            "right": "issue:45092",
+            "accept": false,
+            "reason": "TF32 API guidance and remote-code meta-init incompatibility are different topics."
+          },
+          {
+            "left": "issue:43475",
+            "right": "issue:43643",
+            "accept": false,
+            "reason": "SAM3 output attribute missing and trust_remote_code missing fields are distinct model-loading issues."
+          },
+          {
+            "left": "issue:45230",
+            "right": "issue:45397",
+            "accept": false,
+            "reason": "Generic bug report and Gemma-4 Zero3 loading failure are not the same underlying problem."
+          },
+          {
+            "left": "issue:44038",
+            "right": "issue:44623",
+            "accept": false,
+            "reason": "Qwen3-VL-Moe from_pretrained and processor.save_pretrained missing files are different regressions."
+          },
+          {
+            "left": "issue:41628",
+            "right": "issue:44933",
+            "accept": false,
+            "reason": "Missing AutoImageProcessor import and a nonexistent image_utils import are different API breakages."
+          },
+          {
+            "left": "issue:45081",
+            "right": "issue:45103",
+            "accept": false,
+            "reason": "Tokenizer regex patch crash and docstring kwargs processing crash are unrelated AttributeErrors."
+          },
+          {
+            "left": "issue:42994",
+            "right": "issue:43525",
+            "accept": false,
+            "reason": "Quantized saving failure and Llama4Config missing pad_token_id are separate model/config bugs."
+          },
+          {
+            "left": "issue:45375",
+            "right": "issue:45478",
+            "accept": true,
+            "reason": "Both point at the Qwen3.5-Moe from_pretrained regression; the missing deepstack_visual_indexes field plausibly explains the reported load failure."
+          }
+        ]
+      },
+      "evaluator_result": {
+        "accept": true,
+        "feedback": "Grounded and conservative overall. The false-edge rejections are well supported by the titles, and the one accepted edge stays cautious: the Qwen3.5-Moe from_pretrained report and the missing deepstack_visual_indexes config-field report plausibly describe the same regression family without overstating certainty. The canonical issue choice is also reasonable given the exact same-title report."
+      },
+      "evaluator_used": true,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": true,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 35,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:30333",
+        "issue:30990",
+        "issue:33290",
+        "issue:35141",
+        "issue:36010",
+        "issue:41628",
+        "issue:42175",
+        "issue:42222",
+        "issue:42673",
+        "issue:42831",
+        "issue:42898",
+        "issue:42994",
+        "issue:43232",
+        "issue:43278",
+        "issue:43377",
+        "issue:43475",
+        "issue:43704",
+        "issue:43723",
+        "issue:43784",
+        "issue:43824",
+        "issue:43874",
+        "issue:43901",
+        "issue:44038",
+        "issue:44060",
+        "issue:44188",
+        "issue:44222",
+        "issue:44242",
+        "issue:44351",
+        "issue:44373",
+        "issue:44496",
+        "issue:44568",
+        "issue:44623",
+        "issue:44661",
+        "issue:44734",
+        "issue:44792",
+        "issue:44964",
+        "issue:45092",
+        "issue:45103",
+        "issue:45125",
+        "issue:45137",
+        "issue:45200",
+        "issue:45310",
+        "issue:45356",
+        "issue:45375",
+        "issue:45561",
+        "issue:45600",
+        "issue:45712",
+        "issue:45812"
+      ],
+      "soft_pairs": [
+        "issue:42673|issue:43704",
+        "issue:43475|issue:44661",
+        "issue:45310|issue:45375",
+        "issue:41628|issue:44351",
+        "issue:42831|issue:44222",
+        "issue:42222|issue:45200",
+        "issue:44222|issue:45310",
+        "issue:42994|issue:43824",
+        "issue:42175|issue:45712",
+        "issue:36010|issue:41628",
+        "issue:43377|issue:44242",
+        "issue:44038|issue:45310",
+        "issue:44496|issue:45561",
+        "issue:45561|issue:45600",
+        "issue:44964|issue:45137",
+        "issue:43901|issue:45561",
+        "issue:43232|issue:44792",
+        "issue:42898|issue:44568",
+        "issue:30990|issue:35141",
+        "issue:43232|issue:44188",
+        "issue:42898|issue:45812",
+        "issue:44188|issue:44373",
+        "issue:43784|issue:43824",
+        "issue:44623|issue:45310",
+        "issue:43278|issue:45137",
+        "issue:44734|issue:45103",
+        "issue:30333|issue:33290",
+        "issue:43723|issue:44568",
+        "issue:42898|issue:45356",
+        "issue:44060|issue:45125",
+        "issue:43874|issue:45092"
+      ],
+      "prepared_review_unit_hash": "ae3ee36c606e26e059a8fa9534d1befe06832dee9070f12cb0b83a8d2742410a",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 31,
+        "serialized_chars": 37038,
+        "estimated_input_tokens": 9260,
+        "estimated_eval_tokens": 18776
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "One soft duplicate pair is a clear exact-match bug report about the same Qwen3ForCausalLM VRAM leak; the rest are different issues spanning unrelated import, tokenizer, multimodal, training, and testing regressions.",
+        "confidence": 0.93,
+        "canonical_issue_reason": "issue:42673 is the better canonical: it is the earlier report of the same Qwen3ForCausalLM multi-dataloader VRAM leak and has substantially more discussion activity.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "issue:42673 is the strongest representative issue in this set because it is a concrete, high-signal bug report with a direct duplicate and good engagement.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:42673",
+            "right": "issue:43704",
+            "accept": true,
+            "reason": "Exact same title and same Qwen3ForCausalLM VRAM leak symptom; clearly the same underlying bug."
+          },
+          {
+            "left": "issue:43475",
+            "right": "issue:44661",
+            "accept": false,
+            "reason": "Different failures: missing Sam3VisionEncoderOutput attribute vs tokenizer mapping/add-new-model logic."
+          },
+          {
+            "left": "issue:45310",
+            "right": "issue:45375",
+            "accept": false,
+            "reason": "Same model family, but one is a from_pretrained error and the other is a missing config field; not the same concrete bug."
+          },
+          {
+            "left": "issue:41628",
+            "right": "issue:44351",
+            "accept": false,
+            "reason": "Both are import errors, but for different symbols and different code paths."
+          },
+          {
+            "left": "issue:42831",
+            "right": "issue:44222",
+            "accept": false,
+            "reason": "One is an accuracy regression, the other is a save_pretrained FP8/moe issue."
+          },
+          {
+            "left": "issue:42222",
+            "right": "issue:45200",
+            "accept": false,
+            "reason": "Different model areas and symptoms: vitpose breakage vs Gemma 4 token-type defaults."
+          },
+          {
+            "left": "issue:44222",
+            "right": "issue:45310",
+            "accept": false,
+            "reason": "Related to Qwen3.5 Moe, but save_pretrained FP8 and from_pretrained loading are distinct failures."
+          },
+          {
+            "left": "issue:42994",
+            "right": "issue:43824",
+            "accept": false,
+            "reason": "Quantized saving failure and missing import for a Qwen2.5 VL class are unrelated."
+          },
+          {
+            "left": "issue:42175",
+            "right": "issue:45712",
+            "accept": false,
+            "reason": "Backend dependency installation issue vs leftover dummy class cleanup; not the same bug."
+          },
+          {
+            "left": "issue:36010",
+            "right": "issue:41628",
+            "accept": false,
+            "reason": "Different missing imports: GenerationMixin vs AutoImageProcessor."
+          },
+          {
+            "left": "issue:43377",
+            "right": "issue:44242",
+            "accept": false,
+            "reason": "MIMI padding-mask mismatch and MoE load-balancing-loss logic are unrelated."
+          },
+          {
+            "left": "issue:44038",
+            "right": "issue:45310",
+            "accept": false,
+            "reason": "Both mention Qwen/Moe, but one is a Transformers 5 compatibility problem and the other is a specific loading error."
+          },
+          {
+            "left": "issue:44496",
+            "right": "issue:45561",
+            "accept": false,
+            "reason": "Model config loading failure vs xdist race in testing utilities; no shared bug."
+          },
+          {
+            "left": "issue:45561",
+            "right": "issue:45600",
+            "accept": false,
+            "reason": "Captured-info race in xdist vs stale auto_mappings references; different failure modes."
+          },
+          {
+            "left": "issue:44964",
+            "right": "issue:45137",
+            "accept": false,
+            "reason": "Multimodal model loading and DeepSpeed ZeRO3 deque error are unrelated."
+          },
+          {
+            "left": "issue:43901",
+            "right": "issue:45561",
+            "accept": false,
+            "reason": "Docs mismatch for TextClassificationPipeline vs test-worker file race."
+          },
+          {
+            "left": "issue:43232",
+            "right": "issue:44792",
+            "accept": false,
+            "reason": "Generation kwargs/sync_gpus bug and janus test failure are unrelated."
+          },
+          {
+            "left": "issue:42898",
+            "right": "issue:44568",
+            "accept": false,
+            "reason": "Both are tokenizer regressions, but they affect different APIs and model-specific behavior."
+          },
+          {
+            "left": "issue:30990",
+            "right": "issue:35141",
+            "accept": false,
+            "reason": "Sentence-transformers loading hang and embedding reinitialization on resize are different bugs."
+          },
+          {
+            "left": "issue:43232",
+            "right": "issue:44188",
+            "accept": false,
+            "reason": "Generation kwarg update issue vs torch.compile attention-kernel divergence; different code paths."
+          },
+          {
+            "left": "issue:42898",
+            "right": "issue:45812",
+            "accept": false,
+            "reason": "Different tokenizer regressions affecting different behaviors and model families."
+          },
+          {
+            "left": "issue:44188",
+            "right": "issue:44373",
+            "accept": false,
+            "reason": "Kernel branching bug vs incorrect docstring; not the same underlying issue."
+          },
+          {
+            "left": "issue:43784",
+            "right": "issue:43824",
+            "accept": false,
+            "reason": "Different import-time failures affecting different modules and symbols."
+          },
+          {
+            "left": "issue:44623",
+            "right": "issue:45310",
+            "accept": false,
+            "reason": "Processor save_pretrained missing files and Qwen3.5 Moe loading error are unrelated."
+          },
+          {
+            "left": "issue:43278",
+            "right": "issue:45137",
+            "accept": false,
+            "reason": "Dtype change on evaluation vs DeepSpeed deque underflow; different bugs."
+          },
+          {
+            "left": "issue:44734",
+            "right": "issue:45103",
+            "accept": false,
+            "reason": "KV cache continuation crash and auto_docstring annotation crash are unrelated."
+          },
+          {
+            "left": "issue:30333",
+            "right": "issue:33290",
+            "accept": false,
+            "reason": "MLFlow reporting on failed jobs vs OOM in deepspeed/adafactor are not the same issue."
+          },
+          {
+            "left": "issue:43723",
+            "right": "issue:44568",
+            "accept": false,
+            "reason": "Tokenizer loading regression vs missing BOS/EOS insertion for a specific tokenizer."
+          },
+          {
+            "left": "issue:42898",
+            "right": "issue:45356",
+            "accept": false,
+            "reason": "General v5 tokenization cleanup change vs Kimi-K2.5 codec regression; different problems."
+          },
+          {
+            "left": "issue:44060",
+            "right": "issue:45125",
+            "accept": false,
+            "reason": "Incorrect tied-weights warning and missing tensor-parallel plan are separate model-infrastructure issues."
+          },
+          {
+            "left": "issue:43874",
+            "right": "issue:45092",
+            "accept": false,
+            "reason": "Missing image-patch helper and remote-code/meta-init incompatibility are different failures."
+          }
+        ]
+      },
+      "evaluator_result": {
+        "accept": true,
+        "feedback": "The conclusions are grounded in the packet. The canonical choice for issue:42673 over issue:43704 is reasonable because they share the exact title, same symptom, and the earlier issue has more discussion. The soft-edge verdicts are mostly conservative and do not overclaim equivalence for loosely related reports."
+      },
+      "evaluator_used": true,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": true,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 36,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:28282",
+        "issue:30333",
+        "issue:34689",
+        "issue:35141",
+        "issue:36331",
+        "issue:41628",
+        "issue:42175",
+        "issue:42757",
+        "issue:42831",
+        "issue:42898",
+        "issue:43232",
+        "issue:43479",
+        "issue:43644",
+        "issue:43723",
+        "issue:43873",
+        "issue:43976",
+        "issue:44079",
+        "issue:44188",
+        "issue:44206",
+        "issue:44222",
+        "issue:44261",
+        "issue:44263",
+        "issue:44295",
+        "issue:44297",
+        "issue:44360",
+        "issue:44373",
+        "issue:44448",
+        "issue:44483",
+        "issue:44484",
+        "issue:44485",
+        "issue:44514",
+        "issue:44623",
+        "issue:44779",
+        "issue:44869",
+        "issue:44871",
+        "issue:44933",
+        "issue:44945",
+        "issue:44964",
+        "issue:44977",
+        "issue:45092",
+        "issue:45103",
+        "issue:45310",
+        "issue:45357",
+        "issue:45405",
+        "issue:45406",
+        "issue:45656",
+        "issue:45712",
+        "issue:45834"
+      ],
+      "soft_pairs": [
+        "issue:42898|issue:44779",
+        "issue:45405|issue:45712",
+        "issue:42898|issue:44448",
+        "issue:44188|issue:44964",
+        "issue:44295|issue:45656",
+        "issue:43479|issue:44079",
+        "issue:44933|issue:45712",
+        "issue:42175|issue:44933",
+        "issue:44222|issue:44297",
+        "issue:43976|issue:44933",
+        "issue:43479|issue:44869",
+        "issue:44297|issue:45656",
+        "issue:44206|issue:45103",
+        "issue:42831|issue:44623",
+        "issue:44222|issue:44623",
+        "issue:41628|issue:42757",
+        "issue:30333|issue:36331",
+        "issue:43232|issue:45834",
+        "issue:30333|issue:34689",
+        "issue:43723|issue:44977",
+        "issue:44222|issue:44484",
+        "issue:43873|issue:44945",
+        "issue:44297|issue:45310",
+        "issue:28282|issue:30333",
+        "issue:44263|issue:45092",
+        "issue:30333|issue:35141",
+        "issue:44263|issue:44360",
+        "issue:44483|issue:45406",
+        "issue:44373|issue:44485",
+        "issue:44514|issue:45103",
+        "issue:44295|issue:44623",
+        "issue:43479|issue:43644",
+        "issue:44261|issue:45092",
+        "issue:45092|issue:45357",
+        "issue:43723|issue:44871"
+      ],
+      "prepared_review_unit_hash": "0d0ebba687cc62327e38e04fa2a985b3bc5ba21063610837179af4d7d0edf95a",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 35,
+        "serialized_chars": 38022,
+        "estimated_input_tokens": 9506,
+        "estimated_eval_tokens": 19268
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "Mostly unrelated issues across import errors, tokenizer/model regressions, trainer behavior, and save/load edge cases. The soft-similarity links are thematic at best; none look like the same underlying bug.",
+        "confidence": 0.91,
+        "canonical_issue_reason": "No single canonical issue: the items span many unrelated subsystems and versions, so a dedup canonical would be misleading.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "No global best issue for dedupe; the closest candidates are only superficially similar, not the same bug.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:42898",
+            "right": "issue:44779",
+            "accept": false,
+            "reason": "Both concern tokenizer behavior in v5, but one is about cleanup spacing and the other is a DeepSeek output regression; different bugs."
+          },
+          {
+            "left": "issue:45405",
+            "right": "issue:45712",
+            "accept": false,
+            "reason": "One is a dependency version bump, the other is cleanup of leftover dummy classes; unrelated changes."
+          },
+          {
+            "left": "issue:42898",
+            "right": "issue:44448",
+            "accept": false,
+            "reason": "Both mention v5 output changes, but they affect different models and code paths."
+          },
+          {
+            "left": "issue:44188",
+            "right": "issue:44964",
+            "accept": false,
+            "reason": "Torch.compile attention divergence and Phi-4 model loading failure are different problems."
+          },
+          {
+            "left": "issue:44295",
+            "right": "issue:45656",
+            "accept": false,
+            "reason": "Buffer registration/read failure and Deepspeed double optimizer step are unrelated."
+          },
+          {
+            "left": "issue:43479",
+            "right": "issue:44079",
+            "accept": false,
+            "reason": "Config default initialization and ModelOutput key assignment are different code paths."
+          },
+          {
+            "left": "issue:44933",
+            "right": "issue:45712",
+            "accept": false,
+            "reason": "Import/export failure from image_utils is not the same as removing dummy PT classes."
+          },
+          {
+            "left": "issue:42175",
+            "right": "issue:44933",
+            "accept": false,
+            "reason": "Packaging backend omission and a missing image_utils import are different issues."
+          },
+          {
+            "left": "issue:44222",
+            "right": "issue:44297",
+            "accept": false,
+            "reason": "Both involve save_pretrained, but one is FP8 MoE saving and the other is tokenizer config mismatch."
+          },
+          {
+            "left": "issue:43976",
+            "right": "issue:44933",
+            "accept": false,
+            "reason": "Python version support regression is unrelated to an import-from-image_utils issue."
+          },
+          {
+            "left": "issue:43479",
+            "right": "issue:44869",
+            "accept": false,
+            "reason": "A multimodal config defaulting bug is not the same as Whisper word timestamp decode crashing."
+          },
+          {
+            "left": "issue:44297",
+            "right": "issue:45656",
+            "accept": false,
+            "reason": "Tokenizer save metadata mismatch and Deepspeed optimizer stepping are unrelated."
+          },
+          {
+            "left": "issue:44206",
+            "right": "issue:45103",
+            "accept": false,
+            "reason": "Feature extractor argument regression and auto-docstring annotation handling are different bugs."
+          },
+          {
+            "left": "issue:42831",
+            "right": "issue:44623",
+            "accept": false,
+            "reason": "FP8 accuracy problems and missing processor save_pretrained files are different failure modes."
+          },
+          {
+            "left": "issue:44222",
+            "right": "issue:44623",
+            "accept": false,
+            "reason": "Both touch save_pretrained, but one is about MoE FP8 serialization and the other about processor artifacts missing."
+          },
+          {
+            "left": "issue:41628",
+            "right": "issue:42757",
+            "accept": false,
+            "reason": "Both are import errors, but they involve different symbols and likely different dependency breakages."
+          },
+          {
+            "left": "issue:30333",
+            "right": "issue:36331",
+            "accept": false,
+            "reason": "MLflow reporting on failed jobs and Trainer.compute_loss signature mismatch are unrelated."
+          },
+          {
+            "left": "issue:43232",
+            "right": "issue:45834",
+            "accept": false,
+            "reason": "Generation kwargs update behavior and Kosmos2.5 OCR index errors are unrelated."
+          },
+          {
+            "left": "issue:30333",
+            "right": "issue:34689",
+            "accept": false,
+            "reason": "MLflow run reporting and a model-loading regression are different issues."
+          },
+          {
+            "left": "issue:43723",
+            "right": "issue:44977",
+            "accept": false,
+            "reason": "Tokenizer loading regression and flash-attention generation issues are not the same bug."
+          },
+          {
+            "left": "issue:44222",
+            "right": "issue:44484",
+            "accept": false,
+            "reason": "A save_pretrained FP8 bug and a max_shard_size question are unrelated."
+          },
+          {
+            "left": "issue:43873",
+            "right": "issue:44945",
+            "accept": false,
+            "reason": "Quantization/offloading behavior and pipeline-parallel output correctness are different problems."
+          },
+          {
+            "left": "issue:44297",
+            "right": "issue:45310",
+            "accept": false,
+            "reason": "Tokenizer save metadata mismatch is unrelated to Qwen3.5 MoE from_pretrained failure."
+          },
+          {
+            "left": "issue:28282",
+            "right": "issue:30333",
+            "accept": false,
+            "reason": "Missing PyTorch import support and MLflow job reporting are unrelated."
+          },
+          {
+            "left": "issue:44263",
+            "right": "issue:45092",
+            "accept": false,
+            "reason": "GlmMoeDsaIndexer split-value behavior and InternVL meta-init incompatibility are different code paths."
+          },
+          {
+            "left": "issue:30333",
+            "right": "issue:35141",
+            "accept": false,
+            "reason": "MLflow reporting on failure and output embedding reinitialization are unrelated."
+          },
+          {
+            "left": "issue:44263",
+            "right": "issue:44360",
+            "accept": false,
+            "reason": "Both mention the DSA indexer, but one is about torch.split return handling and the other about a missing ReLU; not the same bug."
+          },
+          {
+            "left": "issue:44483",
+            "right": "issue:45406",
+            "accept": false,
+            "reason": "Chat completions request acceptance and Gemma4Processor missing _tokenizer are different failures."
+          },
+          {
+            "left": "issue:44373",
+            "right": "issue:44485",
+            "accept": false,
+            "reason": "Wrong docstring and GLM-5 RoPE implementation are not the same underlying issue."
+          },
+          {
+            "left": "issue:44514",
+            "right": "issue:45103",
+            "accept": false,
+            "reason": "Batched chat-template crash and future-annotations auto-docstring crash are unrelated."
+          },
+          {
+            "left": "issue:44295",
+            "right": "issue:45092",
+            "accept": false,
+            "reason": "Reading a registered buffer and remote-code meta initialization incompatibility are different problems."
+          },
+          {
+            "left": "issue:45092",
+            "right": "issue:45357",
+            "accept": false,
+            "reason": "Both are version-related model regressions, but they affect different models and different save/load paths."
+          },
+          {
+            "left": "issue:43723",
+            "right": "issue:44871",
+            "accept": false,
+            "reason": "Tokenizer loading regression and eos_token_id config inconsistency are different issues."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 37,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:22355",
+        "issue:28282",
+        "issue:30990",
+        "issue:33357",
+        "issue:38175",
+        "issue:39401",
+        "issue:41628",
+        "issue:42371",
+        "issue:42757",
+        "issue:42831",
+        "issue:42886",
+        "issue:42907",
+        "issue:42994",
+        "issue:43066",
+        "issue:43122",
+        "issue:43232",
+        "issue:43408",
+        "issue:43475",
+        "issue:43479",
+        "issue:43576",
+        "issue:43723",
+        "issue:43756",
+        "issue:43784",
+        "issue:43824",
+        "issue:43937",
+        "issue:44038",
+        "issue:44060",
+        "issue:44117",
+        "issue:44261",
+        "issue:44263",
+        "issue:44295",
+        "issue:44297",
+        "issue:44373",
+        "issue:44442",
+        "issue:44514",
+        "issue:44568",
+        "issue:44704",
+        "issue:44779",
+        "issue:44913",
+        "issue:44945",
+        "issue:45245",
+        "issue:45278",
+        "issue:45290",
+        "issue:45310",
+        "issue:45593",
+        "issue:45600",
+        "issue:45753"
+      ],
+      "soft_pairs": [
+        "issue:30990|issue:38175",
+        "issue:41628|issue:43824",
+        "issue:43479|issue:44913",
+        "issue:42886|issue:42994",
+        "issue:39401|issue:44568",
+        "issue:42831|issue:45310",
+        "issue:22355|issue:28282",
+        "issue:44263|issue:44373",
+        "issue:44945|issue:45753",
+        "issue:42831|issue:44038",
+        "issue:42757|issue:45278",
+        "issue:43576|issue:43784",
+        "issue:42886|issue:43066",
+        "issue:43232|issue:45593",
+        "issue:42371|issue:44263",
+        "issue:43756|issue:44060",
+        "issue:44117|issue:44704",
+        "issue:30990|issue:33357",
+        "issue:42907|issue:44945",
+        "issue:44704|issue:45290",
+        "issue:43122|issue:44514",
+        "issue:44295|issue:44297",
+        "issue:43232|issue:45245",
+        "issue:43723|issue:44442",
+        "issue:43784|issue:45600",
+        "issue:43723|issue:44779",
+        "issue:41628|issue:45278",
+        "issue:42907|issue:43937",
+        "issue:44261|issue:44373",
+        "issue:43408|issue:43475"
+      ],
+      "prepared_review_unit_hash": "eede6dedc36343b6f3259edee00d5993b2cae0dc09d904cba39772aa2ea941a3",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 47,
+        "item_count": 47,
+        "soft_pair_count": 30,
+        "serialized_chars": 35654,
+        "estimated_input_tokens": 8914,
+        "estimated_eval_tokens": 18084
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "These items are mostly separate bug reports with only broad thematic overlap (imports, tokenization, config, generation, quantization). None of the soft pairs look like true duplicates of the same concrete bug/change. If one issue has to represent the set, 45278 is the broadest umbrella, but the cluster is too heterogeneous for a strong canonical duplicate.",
+        "confidence": 0.78,
+        "canonical_issue_reason": "issue:45278 is the broadest representative because it covers multiple import regressions after the 4.57.0\u21925.5.0 upgrade, but it is only a coarse umbrella and not a clean duplicate of the rest.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "issue:45278 is the best single representative for the cluster\u2019s loose import-regression theme, though the underlying reports remain largely distinct.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:30990",
+            "right": "issue:38175",
+            "accept": false,
+            "reason": "Sentence-Transformers loading hang vs. SigLIP2 zero-probability output; different symptoms and codepaths."
+          },
+          {
+            "left": "issue:41628",
+            "right": "issue:43824",
+            "accept": false,
+            "reason": "Missing AutoImageProcessor import vs. missing Qwen2_5_VLForConditionalGeneration import; different symbols and failures."
+          },
+          {
+            "left": "issue:43479",
+            "right": "issue:44913",
+            "accept": false,
+            "reason": "Phi4Multimodal default config initialization vs. GPTNeoX rotary_pct reload behavior; unrelated config bugs."
+          },
+          {
+            "left": "issue:42886",
+            "right": "issue:42994",
+            "accept": false,
+            "reason": "Offline tokenizer cache loading vs. quantized model saving failure; not the same bug."
+          },
+          {
+            "left": "issue:39401",
+            "right": "issue:44568",
+            "accept": false,
+            "reason": "Wrong offset_mapping vs. add_special_tokens not adding BOS/EOS; different tokenizer behaviors."
+          },
+          {
+            "left": "issue:42831",
+            "right": "issue:45310",
+            "accept": false,
+            "reason": "FineGrainedFP8 accuracy issue vs. Qwen3.5 MoE from_pretrained error; different problems."
+          },
+          {
+            "left": "issue:22355",
+            "right": "issue:28282",
+            "accept": false,
+            "reason": "transformers.onnx module missing vs. PyTorch dependency import error; different import failures."
+          },
+          {
+            "left": "issue:44263",
+            "right": "issue:44373",
+            "accept": false,
+            "reason": "torch.split return value issue vs. wrong docstring; one is a code bug, the other documentation."
+          },
+          {
+            "left": "issue:44945",
+            "right": "issue:45753",
+            "accept": false,
+            "reason": "Pipeline-parallelism output bug vs. Qwen3_5 image-specific infinite loop; unrelated inference failures."
+          },
+          {
+            "left": "issue:42831",
+            "right": "issue:44038",
+            "accept": false,
+            "reason": "FineGrainedFP8 accuracy regression vs. Qwen3-VL-Moe bug; different model/codepaths."
+          },
+          {
+            "left": "issue:42757",
+            "right": "issue:45278",
+            "accept": false,
+            "reason": "Specific huggingface_hub API import error vs. broad set of import errors after upgrade; overlap is too vague for a duplicate."
+          },
+          {
+            "left": "issue:43576",
+            "right": "issue:43784",
+            "accept": false,
+            "reason": "Broken transformers env CLI vs. sentence-transformers import NameError; unrelated failure modes."
+          },
+          {
+            "left": "issue:42886",
+            "right": "issue:43066",
+            "accept": false,
+            "reason": "Offline cache loading issue vs. wrong tokenizer decoder type in v5; not the same underlying bug."
+          },
+          {
+            "left": "issue:43232",
+            "right": "issue:45593",
+            "accept": false,
+            "reason": "Generation after sync_gpus issue vs. D-FINE auxiliary-loss handling; unrelated features."
+          },
+          {
+            "left": "issue:42371",
+            "right": "issue:44263",
+            "accept": false,
+            "reason": "TF32 settings warning vs. torch.split return values; no shared bug."
+          },
+          {
+            "left": "issue:43756",
+            "right": "issue:44060",
+            "accept": false,
+            "reason": "Smollm3 RoPE layer count mismatch vs. Qwen3-Next tied-weights warning; different models and defects."
+          },
+          {
+            "left": "issue:44117",
+            "right": "issue:44704",
+            "accept": false,
+            "reason": "TOKENIZER_MAPPING_NAMES None in from_pretrained vs. AutoProcessor kwargs forwarding; different loading paths."
+          },
+          {
+            "left": "issue:30990",
+            "right": "issue:33357",
+            "accept": false,
+            "reason": "Loading hang vs. MacOS bus error; distinct runtime failures."
+          },
+          {
+            "left": "issue:42907",
+            "right": "issue:44945",
+            "accept": false,
+            "reason": "Saving dequantized models vs. incorrect pipeline-parallel output; unrelated codepaths."
+          },
+          {
+            "left": "issue:44704",
+            "right": "issue:45290",
+            "accept": false,
+            "reason": "AutoProcessor kwargs passthrough vs. chat template crash with tool calls; different APIs and bugs."
+          },
+          {
+            "left": "issue:43122",
+            "right": "issue:44514",
+            "accept": false,
+            "reason": "General v4\u2192v5 tokenization change vs. Qwen2_5_VL chat template batched crash; not the same issue."
+          },
+          {
+            "left": "issue:44295",
+            "right": "issue:44297",
+            "accept": false,
+            "reason": "position_ids buffer read error vs. tokenizer_class mismatch on save_pretrained; different serialization issues."
+          },
+          {
+            "left": "issue:43232",
+            "right": "issue:45245",
+            "accept": false,
+            "reason": "Generation update bug vs. category-count limit runtime error; unrelated."
+          },
+          {
+            "left": "issue:43723",
+            "right": "issue:44442",
+            "accept": false,
+            "reason": "Generic AutoTokenizer loading regression vs. failure for a specific tokenizer class; not enough evidence of the same bug."
+          },
+          {
+            "left": "issue:43784",
+            "right": "issue:45600",
+            "accept": false,
+            "reason": "sentence-transformers import NameError vs. Sam3LiteText config references breaking CI; unrelated."
+          },
+          {
+            "left": "issue:43723",
+            "right": "issue:44779",
+            "accept": false,
+            "reason": "Tokenizer loading issue vs. wrong Deepseek tokenization output; different failure classes."
+          },
+          {
+            "left": "issue:41628",
+            "right": "issue:45278",
+            "accept": false,
+            "reason": "Specific AutoImageProcessor import error vs. broad import-error umbrella; too general to treat as one duplicate bug."
+          },
+          {
+            "left": "issue:42907",
+            "right": "issue:43937",
+            "accept": false,
+            "reason": "Quantized model saving failure vs. invalid GenerationConfig in GLM-5; unrelated."
+          },
+          {
+            "left": "issue:44261",
+            "right": "issue:44373",
+            "accept": false,
+            "reason": "MLA precision/config issue vs. docstring typo; clearly not duplicates."
+          },
+          {
+            "left": "issue:43408",
+            "right": "issue:43475",
+            "accept": false,
+            "reason": "SAM3 model-type warning vs. missing fpn_position_embeddings attribute; related area but different concrete bugs."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 38,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:22355",
+        "issue:28218",
+        "issue:29127",
+        "issue:30990",
+        "issue:36296",
+        "issue:39401",
+        "issue:42222",
+        "issue:42371",
+        "issue:42757",
+        "issue:42831",
+        "issue:42907",
+        "issue:43066",
+        "issue:43116",
+        "issue:43122",
+        "issue:43441",
+        "issue:43452",
+        "issue:43493",
+        "issue:43784",
+        "issue:43824",
+        "issue:43825",
+        "issue:43874",
+        "issue:43937",
+        "issue:43976",
+        "issue:43992",
+        "issue:44038",
+        "issue:44075",
+        "issue:44222",
+        "issue:44295",
+        "issue:44297",
+        "issue:44351",
+        "issue:44360",
+        "issue:44485",
+        "issue:44488",
+        "issue:44521",
+        "issue:44625",
+        "issue:44704",
+        "issue:44779",
+        "issue:44829",
+        "issue:44936",
+        "issue:44964",
+        "issue:44977",
+        "issue:45200",
+        "issue:45216",
+        "issue:45446",
+        "issue:45478",
+        "issue:45563",
+        "issue:45753"
+      ],
+      "soft_pairs": [
+        "issue:44360|issue:44485",
+        "issue:43122|issue:44521",
+        "issue:44625|issue:44829",
+        "issue:43116|issue:45200",
+        "issue:29127|issue:30990",
+        "issue:42907|issue:44964",
+        "issue:43122|issue:44779",
+        "issue:44038|issue:44297",
+        "issue:42907|issue:43452",
+        "issue:44038|issue:44295",
+        "issue:39401|issue:44488",
+        "issue:43824|issue:44351",
+        "issue:36296|issue:43116",
+        "issue:44038|issue:44222",
+        "issue:39401|issue:44521",
+        "issue:45216|issue:45478",
+        "issue:42757|issue:44351",
+        "issue:43066|issue:44779",
+        "issue:22355|issue:28218",
+        "issue:42907|issue:44222",
+        "issue:43122|issue:44977",
+        "issue:43992|issue:44704",
+        "issue:43784|issue:45446",
+        "issue:43937|issue:44485",
+        "issue:42222|issue:43441",
+        "issue:42371|issue:43874",
+        "issue:43976|issue:45446",
+        "issue:43493|issue:45753",
+        "issue:42831|issue:45753",
+        "issue:44075|issue:44936",
+        "issue:44038|issue:45478",
+        "issue:43116|issue:43825",
+        "issue:44222|issue:45478",
+        "issue:43116|issue:43493",
+        "issue:43937|issue:45563"
+      ],
+      "prepared_review_unit_hash": "8c6fe6ef11874691f36e3691ed0a24f388a3072f71ed64d1aa429cd311d82105",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 47,
+        "item_count": 47,
+        "soft_pair_count": 35,
+        "serialized_chars": 37059,
+        "estimated_input_tokens": 9265,
+        "estimated_eval_tokens": 18786
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "The items are mostly unrelated standalone bugs. The soft pairs share broad keywords (tokenizer, save_pretrained, Qwen/GLM, etc.) but do not describe the same concrete defect or fix path, so none should be merged as duplicates.",
+        "confidence": 0.88,
+        "canonical_issue_reason": "issue:43122 is the broadest tokenizer-regression report in the set and one of the few nodes that plausibly sits near several tokenizer-related soft hits, making it the best single anchor if a canonical issue must be chosen.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "Despite the weak cohesion of the cluster, issue:43122 is the most reusable representative because it captures the shared tokenizer-regression theme more generically than the other candidates.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:44360",
+            "right": "issue:44485",
+            "accept": false,
+            "reason": "Different subsystems and failures: DSA/ReLU discussion vs GLM-5 RoPE implementation."
+          },
+          {
+            "left": "issue:43122",
+            "right": "issue:44521",
+            "accept": false,
+            "reason": "Both mention tokenization-adjacent behavior, but one is a version-to-version tokenizer regression and the other is a multimodal chat-template assistant_mask bug."
+          },
+          {
+            "left": "issue:44625",
+            "right": "issue:44829",
+            "accept": false,
+            "reason": "Qwen3.5 num_labels propagation and FlashAttention-3 training degeneration are different code paths and symptoms."
+          },
+          {
+            "left": "issue:43116",
+            "right": "issue:45200",
+            "accept": false,
+            "reason": "Example-script multi-label output issue vs Gemma 4 mm_token_type_ids defaulting problem; not the same bug."
+          },
+          {
+            "left": "issue:29127",
+            "right": "issue:30990",
+            "accept": false,
+            "reason": "Clarified error messaging in layoutlmv3 is unrelated to Sentence Transformers hanging on load."
+          },
+          {
+            "left": "issue:42907",
+            "right": "issue:44964",
+            "accept": false,
+            "reason": "Dequantized save failure for Ministral/Devstral is unrelated to Phi-4 multimodal model loading."
+          },
+          {
+            "left": "issue:43122",
+            "right": "issue:44779",
+            "accept": false,
+            "reason": "Both are tokenizer regressions, but they describe different model-specific behavior and no shared concrete failing path is evident."
+          },
+          {
+            "left": "issue:44038",
+            "right": "issue:44297",
+            "accept": false,
+            "reason": "A Qwen3-VL-Moe load/runtime issue is unrelated to tokenizer_class metadata written by save_pretrained."
+          },
+          {
+            "left": "issue:42907",
+            "right": "issue:43452",
+            "accept": false,
+            "reason": "Saving dequantized models and loading via gguf_file are different code paths and failure modes."
+          },
+          {
+            "left": "issue:44038",
+            "right": "issue:44295",
+            "accept": false,
+            "reason": "Qwen3-VL-Moe failure and position_ids buffer access error are distinct bugs."
+          },
+          {
+            "left": "issue:39401",
+            "right": "issue:44488",
+            "accept": false,
+            "reason": "Wrong offset_mapping in Qwen3 tokenizer is unrelated to failing to load a specific model."
+          },
+          {
+            "left": "issue:43824",
+            "right": "issue:44351",
+            "accept": false,
+            "reason": "Both are import errors, but they concern different missing symbols from different packages."
+          },
+          {
+            "left": "issue:36296",
+            "right": "issue:43116",
+            "accept": false,
+            "reason": "Tensor-parallel training bug and empty multi-label example output are unrelated."
+          },
+          {
+            "left": "issue:44038",
+            "right": "issue:44222",
+            "accept": false,
+            "reason": "Qwen3-VL-Moe loading/version problem is not the same as FP8 save_pretrained for MoE."
+          },
+          {
+            "left": "issue:39401",
+            "right": "issue:44521",
+            "accept": false,
+            "reason": "Wrong tokenizer offset_mapping and zero assistant_masks are separate tokenizer-related bugs."
+          },
+          {
+            "left": "issue:45216",
+            "right": "issue:45478",
+            "accept": false,
+            "reason": "One is a save_pretrained regression and the other is a from_pretrained regression; related model family, but not the same concrete failure."
+          },
+          {
+            "left": "issue:42757",
+            "right": "issue:44351",
+            "accept": false,
+            "reason": "Missing is_offline_mode from huggingface_hub and missing HybridCache from transformers are different import/export issues."
+          },
+          {
+            "left": "issue:43066",
+            "right": "issue:44779",
+            "accept": false,
+            "reason": "Wrong tokenizer decoder type in v5 and Deepseek tokenization regression are different tokenizer-path problems."
+          },
+          {
+            "left": "issue:22355",
+            "right": "issue:28218",
+            "accept": false,
+            "reason": "ONNX module import failure is unrelated to added-token spacing behavior."
+          },
+          {
+            "left": "issue:42907",
+            "right": "issue:44222",
+            "accept": false,
+            "reason": "Both involve saving models, but one is dequantized Ministral/Devstral saving and the other is FP8 MoE saving."
+          },
+          {
+            "left": "issue:43122",
+            "right": "issue:44977",
+            "accept": false,
+            "reason": "Tokenizer regression and flash-attention generation failure are unrelated."
+          },
+          {
+            "left": "issue:43992",
+            "right": "issue:44704",
+            "accept": false,
+            "reason": "Missing embed_tokens.weight in UMT5Encoder is not the same as AutoProcessor dropping kwargs to cached_file."
+          },
+          {
+            "left": "issue:43784",
+            "right": "issue:45446",
+            "accept": false,
+            "reason": "sentence-transformers import NameError and flex_attention PyTorch version check are unrelated import/version bugs."
+          },
+          {
+            "left": "issue:43937",
+            "right": "issue:44485",
+            "accept": false,
+            "reason": "Invalid GenerationConfig and GLM-5 RoPE implementation concern different failure modes."
+          },
+          {
+            "left": "issue:42222",
+            "right": "issue:43441",
+            "accept": false,
+            "reason": "Broken vitpose model imports and Ministral flash-attention failure do not share a concrete code path."
+          },
+          {
+            "left": "issue:42371",
+            "right": "issue:43874",
+            "accept": false,
+            "reason": "TF32 API guidance and missing get_number_of_image_patches are unrelated."
+          },
+          {
+            "left": "issue:43976",
+            "right": "issue:45446",
+            "accept": false,
+            "reason": "Python version compatibility and a torch-import version check are different compatibility problems."
+          },
+          {
+            "left": "issue:43493",
+            "right": "issue:45753",
+            "accept": false,
+            "reason": "SigLIP2 implementation discrepancy and a Qwen3_5 image-specific infinite loop are distinct bugs."
+          },
+          {
+            "left": "issue:42831",
+            "right": "issue:45753",
+            "accept": false,
+            "reason": "FP8 accuracy degradation is unrelated to an image-triggered generation loop."
+          },
+          {
+            "left": "issue:44075",
+            "right": "issue:44936",
+            "accept": false,
+            "reason": "SGD arguments being ignored and trainer.evaluate failing after training are separate training issues."
+          },
+          {
+            "left": "issue:44038",
+            "right": "issue:45478",
+            "accept": false,
+            "reason": "Different Qwen3 family variants and different failure surfaces; not the same concrete bug."
+          },
+          {
+            "left": "issue:43116",
+            "right": "issue:43825",
+            "accept": false,
+            "reason": "Example-script classification output bug is unrelated to the pipeline translation-task warning message."
+          },
+          {
+            "left": "issue:43116",
+            "right": "issue:43493",
+            "accept": false,
+            "reason": "Multi-label classification output issue and SigLIP2 implementation discrepancy are unrelated."
+          },
+          {
+            "left": "issue:43937",
+            "right": "issue:45563",
+            "accept": false,
+            "reason": "GenerationConfig validation failure is unrelated to a stale warning in paged generate()."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 39,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:22355",
+        "issue:29127",
+        "issue:33453",
+        "issue:38617",
+        "issue:39401",
+        "issue:41628",
+        "issue:42175",
+        "issue:42222",
+        "issue:42673",
+        "issue:43012",
+        "issue:43064",
+        "issue:43066",
+        "issue:43122",
+        "issue:43381",
+        "issue:43408",
+        "issue:43421",
+        "issue:43450",
+        "issue:43577",
+        "issue:43673",
+        "issue:43742",
+        "issue:43827",
+        "issue:43937",
+        "issue:43986",
+        "issue:44075",
+        "issue:44117",
+        "issue:44291",
+        "issue:44410",
+        "issue:44483",
+        "issue:44484",
+        "issue:44496",
+        "issue:44779",
+        "issue:44829",
+        "issue:44945",
+        "issue:44964",
+        "issue:44977",
+        "issue:45071",
+        "issue:45072",
+        "issue:45081",
+        "issue:45083",
+        "issue:45216",
+        "issue:45278",
+        "issue:45310",
+        "issue:45405",
+        "issue:45446",
+        "issue:45464",
+        "issue:45542",
+        "issue:45588",
+        "issue:45656"
+      ],
+      "soft_pairs": [
+        "issue:43064|issue:43673",
+        "issue:45083|issue:45446",
+        "issue:38617|issue:45278",
+        "issue:33453|issue:39401",
+        "issue:39401|issue:44779",
+        "issue:22355|issue:29127",
+        "issue:44117|issue:45588",
+        "issue:43827|issue:45446",
+        "issue:43381|issue:44945",
+        "issue:43937|issue:44977",
+        "issue:45405|issue:45446",
+        "issue:43450|issue:43986",
+        "issue:42673|issue:45072",
+        "issue:44117|issue:45081",
+        "issue:38617|issue:41628",
+        "issue:39401|issue:43421",
+        "issue:44410|issue:45310",
+        "issue:44829|issue:45216",
+        "issue:44291|issue:45071",
+        "issue:44484|issue:45310",
+        "issue:42222|issue:45542",
+        "issue:43408|issue:43742",
+        "issue:43937|issue:44496",
+        "issue:42175|issue:42222",
+        "issue:44483|issue:45464",
+        "issue:44075|issue:45656",
+        "issue:44964|issue:45278",
+        "issue:43012|issue:43577",
+        "issue:43066|issue:43122"
+      ],
+      "prepared_review_unit_hash": "7a700f14098d723ef94248f3f0db46bd35dba2d2640c4192995fc39e36ecebbf",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 29,
+        "serialized_chars": 36363,
+        "estimated_input_tokens": 9091,
+        "estimated_eval_tokens": 18438
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "The cluster is a collection of unrelated Transformers issues with some title-level similarity, but none of the soft edges look like true duplicates. I rejected all pairings and found no safe canonical issue/PR representative.",
+        "confidence": 0.88,
+        "canonical_issue_reason": "No canonical issue: the items span many different bug classes and model-specific regressions, so none is a reliable duplicate target for the whole cluster.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "No best issue: there is no single issue that meaningfully represents the cluster, since the candidates are about distinct subsystems, models, and failure modes.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:43064",
+            "right": "issue:43673",
+            "accept": false,
+            "reason": "Different code paths: distributed training optimizer-state corruption vs generation cache missing in chunked prefill."
+          },
+          {
+            "left": "issue:45083",
+            "right": "issue:45446",
+            "accept": false,
+            "reason": "Unrelated bugs: qwen3_omni_moe feature behavior vs a flex_attention PyTorch version check."
+          },
+          {
+            "left": "issue:38617",
+            "right": "issue:45278",
+            "accept": false,
+            "reason": "One is a specific missing import symbol; the other is a broad report of many import errors after upgrade."
+          },
+          {
+            "left": "issue:33453",
+            "right": "issue:39401",
+            "accept": false,
+            "reason": "Tokenizer regression vs wrong offset_mapping for Qwen3; related domain but different concrete failures."
+          },
+          {
+            "left": "issue:39401",
+            "right": "issue:44779",
+            "accept": false,
+            "reason": "Different model-specific tokenizer bugs: offset_mapping error vs incorrect DeepSeek tokenization."
+          },
+          {
+            "left": "issue:22355",
+            "right": "issue:29127",
+            "accept": false,
+            "reason": "Completely different problems: missing transformers.onnx module vs a LayoutLMv3 error message issue."
+          },
+          {
+            "left": "issue:44117",
+            "right": "issue:45588",
+            "accept": false,
+            "reason": "Tokenizer mapping lookup failure vs flash_attention crash on s_aux=None; not the same bug."
+          },
+          {
+            "left": "issue:43827",
+            "right": "issue:45446",
+            "accept": false,
+            "reason": "Docs referencing removed pipeline() vs an attention version-check bug; no shared code-path defect."
+          },
+          {
+            "left": "issue:43381",
+            "right": "issue:44945",
+            "accept": false,
+            "reason": "Gradient checkpointing in eval mode and pipeline parallelism output errors are distinct failures."
+          },
+          {
+            "left": "issue:43937",
+            "right": "issue:44977",
+            "accept": false,
+            "reason": "Different model regressions: invalid GenerationConfig for GLM-5 vs Qwen3.5 flash-attention generation failure."
+          },
+          {
+            "left": "issue:45405",
+            "right": "issue:45446",
+            "accept": false,
+            "reason": "PEFT dependency version mismatch is unrelated to the flex_attention import/version-check problem."
+          },
+          {
+            "left": "issue:43450",
+            "right": "issue:43986",
+            "accept": false,
+            "reason": "Video batch shape handling vs AutoProcessor crash without torchvision; different root causes."
+          },
+          {
+            "left": "issue:42673",
+            "right": "issue:45072",
+            "accept": false,
+            "reason": "VRAM leak in Qwen3ForCausalLM multithreaded loading vs dtype mismatch in CI; unrelated."
+          },
+          {
+            "left": "issue:44117",
+            "right": "issue:45081",
+            "accept": false,
+            "reason": "Both are tokenizer loading issues, but one is mapping lookup and the other is a Mistral regex patch crash."
+          },
+          {
+            "left": "issue:38617",
+            "right": "issue:41628",
+            "accept": false,
+            "reason": "Different missing-import regressions for different symbols/modules, not one underlying bug."
+          },
+          {
+            "left": "issue:39401",
+            "right": "issue:43421",
+            "accept": false,
+            "reason": "Offset mapping bug and runtime post-processor update behavior are separate tokenizer issues."
+          },
+          {
+            "left": "issue:44410",
+            "right": "issue:45310",
+            "accept": false,
+            "reason": "Missing projections in qwen3next layer 0 vs Qwen3.5 MoE from_pretrained error; not the same defect."
+          },
+          {
+            "left": "issue:44829",
+            "right": "issue:45216",
+            "accept": false,
+            "reason": "Degenerate training with flash_attention_3 is unrelated to a wrong saved checkpoint from save_pretrained()."
+          },
+          {
+            "left": "issue:44291",
+            "right": "issue:45071",
+            "accept": false,
+            "reason": "Unexpected _is_hf_initialized arg during init_empty_weights vs PretrainedConfig type checking; different failures."
+          },
+          {
+            "left": "issue:44484",
+            "right": "issue:45310",
+            "accept": false,
+            "reason": "A question about max_shard_size is not the same as a Qwen3.5 MoE loading error."
+          },
+          {
+            "left": "issue:42222",
+            "right": "issue:45542",
+            "accept": false,
+            "reason": "VitPose model breakage and TensorBoard-without-TF backend errors are unrelated."
+          },
+          {
+            "left": "issue:43408",
+            "right": "issue:43742",
+            "accept": false,
+            "reason": "Model-type warning for sam3_tracker vs key error loading MobileLLM-125M; different issues."
+          },
+          {
+            "left": "issue:43937",
+            "right": "issue:44496",
+            "accept": false,
+            "reason": "Two different model loading problems: invalid generation config vs unrecognized model/config missing model_type."
+          },
+          {
+            "left": "issue:42175",
+            "right": "issue:42222",
+            "accept": false,
+            "reason": "Dependency installation/backends issue vs broken VitPose models; no shared underlying bug."
+          },
+          {
+            "left": "issue:44483",
+            "right": "issue:45464",
+            "accept": false,
+            "reason": "Both mention chat/completions, but one is a critical API request rejection and the other is streaming inference failure on a specific model."
+          },
+          {
+            "left": "issue:44075",
+            "right": "issue:45656",
+            "accept": false,
+            "reason": "Optimizer args being ignored is different from optimizer.step() being called twice under deepspeed."
+          },
+          {
+            "left": "issue:44964",
+            "right": "issue:45278",
+            "accept": false,
+            "reason": "Specific Phi-4 multimodal loading failure vs a general report of many import errors after upgrade."
+          },
+          {
+            "left": "issue:43012",
+            "right": "issue:43577",
+            "accept": false,
+            "reason": "Both involve dtype/precision, but one is a compile-time bf16 warning and the other is incorrect model dtype after loading Blip2."
+          },
+          {
+            "left": "issue:43066",
+            "right": "issue:43122",
+            "accept": false,
+            "reason": "Tokenizer decoder type mismatch and changed tokenization output are related to v5, but not clearly the same concrete bug."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 40,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:38617",
+        "issue:41628",
+        "issue:42222",
+        "issue:42371",
+        "issue:42673",
+        "issue:42757",
+        "issue:42831",
+        "issue:43012",
+        "issue:43064",
+        "issue:43097",
+        "issue:43335",
+        "issue:43408",
+        "issue:43441",
+        "issue:43452",
+        "issue:43577",
+        "issue:43874",
+        "issue:43906",
+        "issue:43931",
+        "issue:43976",
+        "issue:43986",
+        "issue:44038",
+        "issue:44062",
+        "issue:44117",
+        "issue:44220",
+        "issue:44261",
+        "issue:44263",
+        "issue:44315",
+        "issue:44351",
+        "issue:44368",
+        "issue:44410",
+        "issue:44484",
+        "issue:44485",
+        "issue:44493",
+        "issue:44704",
+        "issue:44829",
+        "issue:44945",
+        "issue:44964",
+        "issue:44977",
+        "issue:44987",
+        "issue:45071",
+        "issue:45103",
+        "issue:45245",
+        "issue:45341",
+        "issue:45375",
+        "issue:45459",
+        "issue:45478",
+        "issue:45542",
+        "issue:45588"
+      ],
+      "soft_pairs": [
+        "issue:42371|issue:44261",
+        "issue:38617|issue:44220",
+        "issue:43874|issue:44261",
+        "issue:44410|issue:45478",
+        "issue:44484|issue:45478",
+        "issue:43097|issue:44493",
+        "issue:44263|issue:44485",
+        "issue:42222|issue:45459",
+        "issue:43012|issue:43408",
+        "issue:43441|issue:44829",
+        "issue:41628|issue:45071",
+        "issue:42371|issue:42831",
+        "issue:45071|issue:45542",
+        "issue:42673|issue:44977",
+        "issue:42673|issue:43906",
+        "issue:43931|issue:44038",
+        "issue:44117|issue:44987",
+        "issue:43097|issue:43452",
+        "issue:43335|issue:45071",
+        "issue:43976|issue:44945",
+        "issue:43408|issue:44368",
+        "issue:43874|issue:44263",
+        "issue:43986|issue:44484",
+        "issue:44829|issue:45245",
+        "issue:38617|issue:42757",
+        "issue:43064|issue:44315",
+        "issue:44410|issue:45375",
+        "issue:38617|issue:44351",
+        "issue:43097|issue:45588",
+        "issue:44062|issue:44117",
+        "issue:44038|issue:44410",
+        "issue:42673|issue:43577",
+        "issue:44964|issue:45071",
+        "issue:44704|issue:45103",
+        "issue:43931|issue:44410",
+        "issue:44261|issue:44485",
+        "issue:44964|issue:45341"
+      ],
+      "prepared_review_unit_hash": "5e824c0973ecd4ad487ffcae8f23bc9e844b2d8ba868bd6ebdfb81c4c5449f5e",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 37,
+        "serialized_chars": 38783,
+        "estimated_input_tokens": 9696,
+        "estimated_eval_tokens": 19648
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "These soft pairs are mostly thematic, not true duplicates: import errors, model-specific bugs, precision warnings, and trainer/config regressions are distinct code paths. No PRs are present. If a single issue must represent the cluster, 44261 is the best fit because it is a concrete, well-scoped numerical bug with multiple nearby soft links.",
+        "confidence": 0.71,
+        "canonical_issue_reason": "issue:44261 is the most substantive and central of the provided issues, with the most soft links and a narrowly described precision/numerics bug; it is a better representative than the more generic import/error reports.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "issue:44261 is the best representative issue overall: it is concrete, reproducible, and closest to multiple adjacent precision-related reports, even though the cluster is not truly duplicate-heavy.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:42371",
+            "right": "issue:44261",
+            "accept": false,
+            "reason": "TF32 backend-setting discussion and MLA rms_norm_eps precision bug are different features and failure modes."
+          },
+          {
+            "left": "issue:38617",
+            "right": "issue:44220",
+            "accept": false,
+            "reason": "Two unrelated errors in different code paths: an import error vs an audio feature extraction issue."
+          },
+          {
+            "left": "issue:43874",
+            "right": "issue:44261",
+            "accept": false,
+            "reason": "Missing image-patch method in a vision processor is unrelated to the MLA layernorm precision/config problem."
+          },
+          {
+            "left": "issue:44410",
+            "right": "issue:45478",
+            "accept": false,
+            "reason": "Different Qwen model families and different failures; not the same concrete bug."
+          },
+          {
+            "left": "issue:44484",
+            "right": "issue:45478",
+            "accept": false,
+            "reason": "Save-pretrained shard sizing is unrelated to Qwen3.5 MoE from_pretrained loading failures."
+          },
+          {
+            "left": "issue:43097",
+            "right": "issue:44493",
+            "accept": false,
+            "reason": "One is an embeddings-API removal issue; the other is a position-id warning. Different regressions."
+          },
+          {
+            "left": "issue:44263",
+            "right": "issue:44485",
+            "accept": false,
+            "reason": "Torch split/indexing behavior and GLM-5 RoPE math are different bugs, not one shared code-path issue."
+          },
+          {
+            "left": "issue:42222",
+            "right": "issue:45459",
+            "accept": false,
+            "reason": "VitPose model breakage and tokenizer protobuf error masking are unrelated subsystems."
+          },
+          {
+            "left": "issue:43012",
+            "right": "issue:43408",
+            "accept": false,
+            "reason": "A PyTorch bfloat16 warning and a SAM3 model-type warning are not the same bug."
+          },
+          {
+            "left": "issue:43441",
+            "right": "issue:44829",
+            "accept": false,
+            "reason": "Both mention flash attention, but they affect different models and symptoms, so they are not mergeable as one PR/bug."
+          },
+          {
+            "left": "issue:41628",
+            "right": "issue:45071",
+            "accept": false,
+            "reason": "AutoImageProcessor import failure and PreTrainedConfig type-checking regression are unrelated."
+          },
+          {
+            "left": "issue:42371",
+            "right": "issue:42831",
+            "accept": false,
+            "reason": "TF32 API guidance and FineGrainedFP8 accuracy problems are distinct precision topics, not the same bug."
+          },
+          {
+            "left": "issue:45071",
+            "right": "issue:45542",
+            "accept": false,
+            "reason": "Type-checking regression and undefined tf backend error are unrelated failure modes."
+          },
+          {
+            "left": "issue:42673",
+            "right": "issue:44977",
+            "accept": false,
+            "reason": "VRAM leakage in dataloader threads and flash-attention generation failure are different Qwen3 issues."
+          },
+          {
+            "left": "issue:42673",
+            "right": "issue:43906",
+            "accept": false,
+            "reason": "No evidence these describe the same underlying bug; one is a memory leak, the other a reproduction of a different issue."
+          },
+          {
+            "left": "issue:43931",
+            "right": "issue:44038",
+            "accept": false,
+            "reason": "Both involve Qwen3-VL, but one is a weight-shape mismatch and the other is a model-loading bug for a different variant."
+          },
+          {
+            "left": "issue:44117",
+            "right": "issue:44987",
+            "accept": false,
+            "reason": "Tokenizer mapping returning None and a physical-intelligence/fast load failure are not the same code-path problem."
+          },
+          {
+            "left": "issue:43097",
+            "right": "issue:43452",
+            "accept": false,
+            "reason": "Embeddings/tie-weights removal and gguf_file loading errors are unrelated."
+          },
+          {
+            "left": "issue:43335",
+            "right": "issue:45071",
+            "accept": false,
+            "reason": "SwitchTransformers sparse-layer config bug and PreTrainedConfig type checking are different areas."
+          },
+          {
+            "left": "issue:43976",
+            "right": "issue:44945",
+            "accept": false,
+            "reason": "Python-version compatibility and pipeline-parallel output corruption are unrelated issues."
+          },
+          {
+            "left": "issue:43408",
+            "right": "issue:44368",
+            "accept": false,
+            "reason": "A model-type mismatch warning and a tie_word_embeddings warning are different config warnings."
+          },
+          {
+            "left": "issue:43874",
+            "right": "issue:44263",
+            "accept": false,
+            "reason": "Vision-processor API omission and torch.split return handling are unrelated."
+          },
+          {
+            "left": "issue:43986",
+            "right": "issue:44484",
+            "accept": false,
+            "reason": "AutoProcessor torchvision dependency crash is unrelated to save_pretrained shard sizing."
+          },
+          {
+            "left": "issue:44829",
+            "right": "issue:45245",
+            "accept": false,
+            "reason": "Flash-attention training degradation and category-count limit are unrelated."
+          },
+          {
+            "left": "issue:38617",
+            "right": "issue:42757",
+            "accept": false,
+            "reason": "Different import errors from different symbols; same symptom class, but not the same bug."
+          },
+          {
+            "left": "issue:43064",
+            "right": "issue:44315",
+            "accept": false,
+            "reason": "Wrong optimizer-state handling in FSDP2/PEFT and Liger Kernel application with model_init are different training bugs."
+          },
+          {
+            "left": "issue:44410",
+            "right": "issue:45375",
+            "accept": false,
+            "reason": "Missing qkv/gate projections and a missing config field in vision MoE are not the same issue."
+          },
+          {
+            "left": "issue:38617",
+            "right": "issue:44351",
+            "accept": false,
+            "reason": "ImportError for layer_type_validation and HybridCache are separate missing-symbol regressions."
+          },
+          {
+            "left": "issue:43097",
+            "right": "issue:45588",
+            "accept": false,
+            "reason": "Embeddings tie-weights API removal and flash_attention.py s_aux handling are unrelated."
+          },
+          {
+            "left": "issue:44062",
+            "right": "issue:44117",
+            "accept": false,
+            "reason": "Tokenizer AddedToken construction error and tokenizer-mapping None handling are different tokenizer bugs."
+          },
+          {
+            "left": "issue:44038",
+            "right": "issue:44410",
+            "accept": false,
+            "reason": "Qwen3-VL-Moe loading/bug and qwen3next missing projections are different model implementations and failure modes."
+          },
+          {
+            "left": "issue:42673",
+            "right": "issue:43577",
+            "accept": false,
+            "reason": "Memory leak in Qwen3ForCausalLM and Blip2 dtype propagation are unrelated."
+          },
+          {
+            "left": "issue:44964",
+            "right": "issue:45071",
+            "accept": false,
+            "reason": "Phi-4 multimodal loading failure and PreTrainedConfig type-checking regression are not the same code-path issue."
+          },
+          {
+            "left": "issue:44704",
+            "right": "issue:45103",
+            "accept": false,
+            "reason": "AutoProcessor kwargs forwarding and auto_docstring annotation handling are unrelated."
+          },
+          {
+            "left": "issue:44261",
+            "right": "issue:44485",
+            "accept": false,
+            "reason": "Layernorm epsilon precision in MLA and GLM-5 RoPE implementation are different numerical bugs."
+          },
+          {
+            "left": "issue:44964",
+            "right": "issue:45341",
+            "accept": false,
+            "reason": "Model loading failure and a testing_utils bug are unrelated."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 41,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:36683",
+        "issue:38617",
+        "issue:42617",
+        "issue:42757",
+        "issue:42886",
+        "issue:42907",
+        "issue:42994",
+        "issue:43012",
+        "issue:43064",
+        "issue:43066",
+        "issue:43097",
+        "issue:43208",
+        "issue:43299",
+        "issue:43381",
+        "issue:43475",
+        "issue:43502",
+        "issue:43526",
+        "issue:43550",
+        "issue:43618",
+        "issue:43761",
+        "issue:43824",
+        "issue:43825",
+        "issue:43867",
+        "issue:43873",
+        "issue:43901",
+        "issue:43906",
+        "issue:43986",
+        "issue:43992",
+        "issue:44038",
+        "issue:44117",
+        "issue:44230",
+        "issue:44368",
+        "issue:44380",
+        "issue:44483",
+        "issue:44484",
+        "issue:44514",
+        "issue:44655",
+        "issue:44683",
+        "issue:45216",
+        "issue:45290",
+        "issue:45341",
+        "issue:45372",
+        "issue:45381",
+        "issue:45561",
+        "issue:45588",
+        "issue:45753",
+        "issue:45859"
+      ],
+      "soft_pairs": [
+        "issue:43502|issue:43992",
+        "issue:43986|issue:44117",
+        "issue:45341|issue:45859",
+        "issue:43097|issue:43873",
+        "issue:43066|issue:43906",
+        "issue:43986|issue:45290",
+        "issue:45341|issue:45561",
+        "issue:43986|issue:45372",
+        "issue:43097|issue:43901",
+        "issue:42617|issue:45341",
+        "issue:43986|issue:45341",
+        "issue:43825|issue:44655",
+        "issue:45341|issue:45588",
+        "issue:43992|issue:44484",
+        "issue:43012|issue:43550",
+        "issue:43208|issue:43550",
+        "issue:43012|issue:44368",
+        "issue:42886|issue:43381",
+        "issue:42907|issue:45753",
+        "issue:42907|issue:44655",
+        "issue:43208|issue:43526",
+        "issue:44483|issue:44514",
+        "issue:43012|issue:44683",
+        "issue:42757|issue:43824",
+        "issue:38617|issue:43502",
+        "issue:44038|issue:44230",
+        "issue:44483|issue:45290",
+        "issue:43618|issue:43761",
+        "issue:43064|issue:44380",
+        "issue:43299|issue:43867",
+        "issue:44483|issue:45381",
+        "issue:43901|issue:44655",
+        "issue:36683|issue:43475",
+        "issue:43867|issue:45216",
+        "issue:42994|issue:44655"
+      ],
+      "prepared_review_unit_hash": "ba1a159d1039fa4457baaca0dcba78c36971ee3b1874dcd87d0a962b88881b7f",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 47,
+        "item_count": 47,
+        "soft_pair_count": 35,
+        "serialized_chars": 37417,
+        "estimated_input_tokens": 9355,
+        "estimated_eval_tokens": 18966
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "This cluster is mostly a set of unrelated Transformers issues that only share high-level themes like imports, loading, saving, quantization, and v5 regressions. I don\u2019t see any soft pair that is clearly the same underlying bug or change.",
+        "confidence": 0.43,
+        "canonical_issue_reason": "issue:43873 is the best umbrella candidate: it is open, has substantial discussion/inbound references, and describes a broad quantization/offloading failure mode that can plausibly collect related reports better than the more narrow, model-specific issues.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "issue:43873 is the strongest global issue choice because it is active, broadly phrased, and sits in a recurring failure area (quantization/offloading) that could absorb adjacent reports if a canonical issue is needed.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:43502",
+            "right": "issue:43992",
+            "accept": false,
+            "reason": "Both involve loading/saving behavior, but one is local_files_only network leakage and the other is a missing weight during UMT5 loading; different code paths and fixes."
+          },
+          {
+            "left": "issue:43986",
+            "right": "issue:44117",
+            "accept": false,
+            "reason": "Different failures: video AutoProcessor setup without torchvision vs tokenizer mapping returning None during from_pretrained."
+          },
+          {
+            "left": "issue:45341",
+            "right": "issue:45859",
+            "accept": false,
+            "reason": "Testing-utils bug versus Qwen3_5Moe CPU-offload incompatibility; unrelated subsystems and symptoms."
+          },
+          {
+            "left": "issue:43097",
+            "right": "issue:43873",
+            "accept": false,
+            "reason": "One is an API removal/documentation issue, the other is a quantization/offloading runtime bug; not the same underlying problem."
+          },
+          {
+            "left": "issue:43066",
+            "right": "issue:43906",
+            "accept": false,
+            "reason": "Wrong tokenizer decoder type and an isolated reproduction of a different issue are not enough to establish the same concrete bug."
+          },
+          {
+            "left": "issue:43986",
+            "right": "issue:45290",
+            "accept": false,
+            "reason": "Both touch multimodal/chat processing, but they fail in different places and with different preconditions."
+          },
+          {
+            "left": "issue:45341",
+            "right": "issue:45561",
+            "accept": false,
+            "reason": "A test utility bug and an xdist file-race are related only by testing infrastructure, not the same defect."
+          },
+          {
+            "left": "issue:43986",
+            "right": "issue:45372",
+            "accept": false,
+            "reason": "Different missing dependency/import problems in different processor/model-loading paths."
+          },
+          {
+            "left": "issue:43097",
+            "right": "issue:43901",
+            "accept": false,
+            "reason": "Removed embedding-tie API versus stale docs about return_all_scores; separate issue classes."
+          },
+          {
+            "left": "issue:42617",
+            "right": "issue:45341",
+            "accept": false,
+            "reason": "A script execution problem and a testing-utils bug are unrelated beyond being bugs in the repo."
+          },
+          {
+            "left": "issue:43986",
+            "right": "issue:45341",
+            "accept": false,
+            "reason": "Video processor import/setup failure versus testing-utils behavior; no shared code path."
+          },
+          {
+            "left": "issue:43825",
+            "right": "issue:44655",
+            "accept": false,
+            "reason": "Pipeline task messaging and pipeline save_pretrained are distinct issues affecting different APIs."
+          },
+          {
+            "left": "issue:45341",
+            "right": "issue:45588",
+            "accept": false,
+            "reason": "Testing utilities and flash_attention sink handling are separate code paths with different failure modes."
+          },
+          {
+            "left": "issue:43992",
+            "right": "issue:44484",
+            "accept": false,
+            "reason": "Missing embed_tokens weight during load is unrelated to the max_shard_size default question in save_pretrained."
+          },
+          {
+            "left": "issue:43012",
+            "right": "issue:43550",
+            "accept": false,
+            "reason": "Precision warning during compile and Bamba SDPA compile failure are different runtime behaviors."
+          },
+          {
+            "left": "issue:43208",
+            "right": "issue:43550",
+            "accept": false,
+            "reason": "xLSTM training bugs versus Bamba torch.compile+SDPA failure; not the same bug."
+          },
+          {
+            "left": "issue:43012",
+            "right": "issue:44368",
+            "accept": false,
+            "reason": "Both mention warnings/config, but one is a PyTorch precision warning and the other is an embedding-tying warning in a different model family."
+          },
+          {
+            "left": "issue:42886",
+            "right": "issue:43381",
+            "accept": false,
+            "reason": "Offline cache loading and gradient-checkpointing-in-eval are unrelated."
+          },
+          {
+            "left": "issue:42907",
+            "right": "issue:45753",
+            "accept": false,
+            "reason": "Saving dequantized models and an infinite-generation loop on a specific image are different defects."
+          },
+          {
+            "left": "issue:42907",
+            "right": "issue:44655",
+            "accept": false,
+            "reason": "Model dequantization save failure is not the same as saving Pipeline objects."
+          },
+          {
+            "left": "issue:43208",
+            "right": "issue:43526",
+            "accept": false,
+            "reason": "xLSTM training issues versus BeitImageProcessorFast label reduction bug; different components."
+          },
+          {
+            "left": "issue:44483",
+            "right": "issue:44514",
+            "accept": false,
+            "reason": "OpenAI-style chat endpoint rejection and Qwen2_5_VLProcessor batching crash are separate request paths."
+          },
+          {
+            "left": "issue:43012",
+            "right": "issue:44683",
+            "accept": false,
+            "reason": "Both are compile-related, but one is a bf16 warning and the other is a flex_attention failure on newer torch versions."
+          },
+          {
+            "left": "issue:42757",
+            "right": "issue:43824",
+            "accept": false,
+            "reason": "Both are import errors, but they involve different missing symbols from different modules and different fixes."
+          },
+          {
+            "left": "issue:38617",
+            "right": "issue:43502",
+            "accept": false,
+            "reason": "Missing layer_type_validation import and local_files_only making API requests are unrelated issues."
+          },
+          {
+            "left": "issue:44038",
+            "right": "issue:44230",
+            "accept": false,
+            "reason": "Qwen3-VL-Moe version breakage and fp8 inference support are not the same concrete bug."
+          },
+          {
+            "left": "issue:44483",
+            "right": "issue:45290",
+            "accept": false,
+            "reason": "Different user-facing APIs: chat completions endpoint handling versus apply_chat_template tokenization crash."
+          },
+          {
+            "left": "issue:43618",
+            "right": "issue:43761",
+            "accept": false,
+            "reason": "Both are CLIP regressions, but one concerns attentions not being assigned and the other hidden_states being None; not the same defect."
+          },
+          {
+            "left": "issue:43064",
+            "right": "issue:44380",
+            "accept": false,
+            "reason": "Trainer/FSDP2 optimizer-state corruption is unrelated to GPT2 attention scaling being ignored under SDPA/FlashAttention."
+          },
+          {
+            "left": "issue:43299",
+            "right": "issue:43867",
+            "accept": false,
+            "reason": "A Qwen3VL-Moe loading regression and a generic state_dict-sorting load error do not look like one underlying bug."
+          },
+          {
+            "left": "issue:44483",
+            "right": "issue:45381",
+            "accept": false,
+            "reason": "Chat-completions API rejection and Qwen2.5-VL video position-id corruption are different paths and symptoms."
+          },
+          {
+            "left": "issue:43901",
+            "right": "issue:44655",
+            "accept": false,
+            "reason": "Documentation drift for return_all_scores does not match a pipeline serialization failure."
+          },
+          {
+            "left": "issue:36683",
+            "right": "issue:43475",
+            "accept": false,
+            "reason": "Gemma3Config missing vocab_size and SAM3VisionEncoderOutput missing fpn_position_embeddings are unrelated attribute errors."
+          },
+          {
+            "left": "issue:43867",
+            "right": "issue:45216",
+            "accept": false,
+            "reason": "Generic state_dict sorting load failure and Qwen3.5 save_pretrained regression are distinct save/load bugs."
+          },
+          {
+            "left": "issue:42994",
+            "right": "issue:44655",
+            "accept": false,
+            "reason": "Quantized model saving failure and saving Pipeline objects are different save_pretrained problems with different object types."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 42,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:36010",
+        "issue:36296",
+        "issue:36683",
+        "issue:41669",
+        "issue:42757",
+        "issue:42890",
+        "issue:43064",
+        "issue:43066",
+        "issue:43097",
+        "issue:43299",
+        "issue:43352",
+        "issue:43441",
+        "issue:43525",
+        "issue:43575",
+        "issue:43618",
+        "issue:43756",
+        "issue:43867",
+        "issue:43881",
+        "issue:43906",
+        "issue:43937",
+        "issue:43976",
+        "issue:43992",
+        "issue:44077",
+        "issue:44117",
+        "issue:44220",
+        "issue:44242",
+        "issue:44246",
+        "issue:44263",
+        "issue:44355",
+        "issue:44360",
+        "issue:44380",
+        "issue:44484",
+        "issue:44485",
+        "issue:44492",
+        "issue:44655",
+        "issue:44855",
+        "issue:44938",
+        "issue:45216",
+        "issue:45310",
+        "issue:45357",
+        "issue:45375",
+        "issue:45381",
+        "issue:45440",
+        "issue:45478",
+        "issue:45479",
+        "issue:45561",
+        "issue:45753"
+      ],
+      "soft_pairs": [
+        "issue:43352|issue:43937",
+        "issue:36010|issue:42757",
+        "issue:43066|issue:45381",
+        "issue:43756|issue:44485",
+        "issue:44220|issue:44655",
+        "issue:44263|issue:45440",
+        "issue:36683|issue:43525",
+        "issue:43906|issue:45753",
+        "issue:43618|issue:45216",
+        "issue:36683|issue:44242",
+        "issue:43299|issue:45753",
+        "issue:36683|issue:43881",
+        "issue:43352|issue:43976",
+        "issue:43441|issue:44380",
+        "issue:43352|issue:44380",
+        "issue:43992|issue:45310",
+        "issue:43867|issue:45375",
+        "issue:44380|issue:45479",
+        "issue:41669|issue:44246",
+        "issue:43352|issue:45478",
+        "issue:43618|issue:44380",
+        "issue:43064|issue:44492",
+        "issue:44355|issue:44855",
+        "issue:36683|issue:44117",
+        "issue:42890|issue:45561",
+        "issue:43097|issue:43352",
+        "issue:43867|issue:45357",
+        "issue:43976|issue:44938",
+        "issue:43992|issue:45478",
+        "issue:44360|issue:45440",
+        "issue:43064|issue:44655",
+        "issue:36296|issue:43575",
+        "issue:44077|issue:44380",
+        "issue:43867|issue:44484"
+      ],
+      "prepared_review_unit_hash": "c8fbfbfe3bae75b9fdde37ca4a045a597112123365f40c942155cb6e65e1686b",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 47,
+        "item_count": 47,
+        "soft_pair_count": 34,
+        "serialized_chars": 36866,
+        "estimated_input_tokens": 9217,
+        "estimated_eval_tokens": 18690
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "This cluster is highly heterogeneous: most items are unrelated model-specific, training, import, or test issues that only look superficially similar by keyword. I would not merge any of the soft pairs as duplicates. The strongest standalone issue here is the import-time performance regression in issue 41669.",
+        "confidence": 0.41,
+        "canonical_issue_reason": "Issue 41669 is the most concrete, actionable, and broadly impactful regression in the set: it describes a measurable import-time slowdown with a plausible single root cause. The rest of the cluster is mostly a mix of unrelated model/config/runtime bugs.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "Issue 41669 is the best anchor for this noisy cluster because it is specific, reproducible, and high-signal compared with the many narrow model-version failures around it.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:43352",
+            "right": "issue:43937",
+            "accept": false,
+            "reason": "Different bugs: FlashAttention support for a model class vs a generation config validation error."
+          },
+          {
+            "left": "issue:36010",
+            "right": "issue:42757",
+            "accept": false,
+            "reason": "Both are import errors, but they involve different packages and different missing symbols."
+          },
+          {
+            "left": "issue:43066",
+            "right": "issue:45381",
+            "accept": false,
+            "reason": "Tokenizer decoder type mismatch and Qwen2.5-VL vision position ids are unrelated failures."
+          },
+          {
+            "left": "issue:43756",
+            "right": "issue:44485",
+            "accept": false,
+            "reason": "Different model families and different problems: RoPE layer count vs RoPE implementation details."
+          },
+          {
+            "left": "issue:44220",
+            "right": "issue:44655",
+            "accept": false,
+            "reason": "Audio feature extraction bug and pipeline save_pretrained behavior are unrelated."
+          },
+          {
+            "left": "issue:44263",
+            "right": "issue:45440",
+            "accept": false,
+            "reason": "GlmMoeDsaIndexer torch.split behavior and DeepseekV3MoE divergence are distinct code paths."
+          },
+          {
+            "left": "issue:36683",
+            "right": "issue:43525",
+            "accept": false,
+            "reason": "Two separate config-attribute omissions on different models; not the same bug."
+          },
+          {
+            "left": "issue:43906",
+            "right": "issue:45753",
+            "accept": false,
+            "reason": "An isolated reproduction of an older issue is not the same as a Qwen3.5 infinite-loop regression."
+          },
+          {
+            "left": "issue:43618",
+            "right": "issue:45216",
+            "accept": false,
+            "reason": "Missing CLIP attentions and incorrect Qwen3.5 checkpoint saving are different regressions."
+          },
+          {
+            "left": "issue:36683",
+            "right": "issue:44242",
+            "accept": false,
+            "reason": "Missing Gemma3Config attribute and load-balancing-loss omission are unrelated."
+          },
+          {
+            "left": "issue:43299",
+            "right": "issue:45753",
+            "accept": false,
+            "reason": "Qwen3VL MoE loading failure and Qwen3_5 image infinite loop are different symptoms and likely different causes."
+          },
+          {
+            "left": "issue:36683",
+            "right": "issue:43881",
+            "accept": false,
+            "reason": "Gemma3Config attribute error and GLM-4V-9B load failure are not the same underlying issue."
+          },
+          {
+            "left": "issue:43352",
+            "right": "issue:43976",
+            "accept": false,
+            "reason": "FlashAttention support bug and Python-version compatibility failure are unrelated."
+          },
+          {
+            "left": "issue:43441",
+            "right": "issue:44380",
+            "accept": false,
+            "reason": "Ministral FlashAttention failure and GPT2 attention-scaling configuration are separate code paths."
+          },
+          {
+            "left": "issue:43352",
+            "right": "issue:44380",
+            "accept": false,
+            "reason": "Different models and different behaviors; one is unsupported FlashAttention, the other is ignored scaling config."
+          },
+          {
+            "left": "issue:43992",
+            "right": "issue:45310",
+            "accept": false,
+            "reason": "UMT5 weight loading and Qwen3.5 MoE from_pretrained failure do not match."
+          },
+          {
+            "left": "issue:43867",
+            "right": "issue:45375",
+            "accept": false,
+            "reason": "State_dict ordering/load error and a missing config field dropped by @strict are not the same bug."
+          },
+          {
+            "left": "issue:44380",
+            "right": "issue:45479",
+            "accept": false,
+            "reason": "Attention scaling in GPT2 and a degenerate zero-loss classification bug are unrelated."
+          },
+          {
+            "left": "issue:41669",
+            "right": "issue:44246",
+            "accept": false,
+            "reason": "Both mention import slowness, but one is a specific import-star regression and the other is a general intermittent slowdown report."
+          },
+          {
+            "left": "issue:43352",
+            "right": "issue:45478",
+            "accept": false,
+            "reason": "Same broad theme, but different models and distinct failures."
+          },
+          {
+            "left": "issue:43618",
+            "right": "issue:44380",
+            "accept": false,
+            "reason": "Missing attentions output and attention scaling being ignored are different issues."
+          },
+          {
+            "left": "issue:43064",
+            "right": "issue:44492",
+            "accept": false,
+            "reason": "A distributed training optimizer-state bug and a cache-strategy typo are not duplicates."
+          },
+          {
+            "left": "issue:44355",
+            "right": "issue:44855",
+            "accept": false,
+            "reason": "Compiled-file import errors and a Python 3.13 decorator/comment parsing failure are different root causes."
+          },
+          {
+            "left": "issue:36683",
+            "right": "issue:44117",
+            "accept": false,
+            "reason": "Config attribute missing and tokenizer mapping returning None are unrelated."
+          },
+          {
+            "left": "issue:42890",
+            "right": "issue:45561",
+            "accept": false,
+            "reason": "Missing set_seed in tests and xdist file-race corruption are both test issues but not the same bug."
+          },
+          {
+            "left": "issue:43097",
+            "right": "issue:43352",
+            "accept": false,
+            "reason": "A removed config field and FlashAttention support are completely different topics."
+          },
+          {
+            "left": "issue:43867",
+            "right": "issue:45357",
+            "accept": false,
+            "reason": "Both involve loading/saving state, but the concrete failure modes differ."
+          },
+          {
+            "left": "issue:43976",
+            "right": "issue:44938",
+            "accept": false,
+            "reason": "These are separate Python-version compatibility reports affecting different releases and environments."
+          },
+          {
+            "left": "issue:43992",
+            "right": "issue:45478",
+            "accept": false,
+            "reason": "Different model families and different missing-weight / from_pretrained failures."
+          },
+          {
+            "left": "issue:44360",
+            "right": "issue:45440",
+            "accept": false,
+            "reason": "DSA indexer math and DeepSeek MoE implementation divergence are unrelated."
+          },
+          {
+            "left": "issue:43064",
+            "right": "issue:44655",
+            "accept": false,
+            "reason": "Distributed optimizer-state corruption and Pipeline save_pretrained support are separate areas."
+          },
+          {
+            "left": "issue:36296",
+            "right": "issue:43575",
+            "accept": false,
+            "reason": "Both mention tensor parallelism, but one is a training bug and the other is a loading/OOM report."
+          },
+          {
+            "left": "issue:44077",
+            "right": "issue:44380",
+            "accept": false,
+            "reason": "PatchTSMixer post_init constraints and GPT2 attention scaling are not the same code-path problem."
+          },
+          {
+            "left": "issue:43867",
+            "right": "issue:44484",
+            "accept": false,
+            "reason": "State_dict load-order failure and the default max_shard_size question are unrelated."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 43,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:28282",
+        "issue:30333",
+        "issue:30990",
+        "issue:31795",
+        "issue:33453",
+        "issue:34689",
+        "issue:36296",
+        "issue:42548",
+        "issue:42757",
+        "issue:42831",
+        "issue:42890",
+        "issue:43010",
+        "issue:43502",
+        "issue:43576",
+        "issue:43618",
+        "issue:43723",
+        "issue:43827",
+        "issue:43874",
+        "issue:43901",
+        "issue:44062",
+        "issue:44075",
+        "issue:44112",
+        "issue:44222",
+        "issue:44230",
+        "issue:44315",
+        "issue:44351",
+        "issue:44355",
+        "issue:44360",
+        "issue:44368",
+        "issue:44387",
+        "issue:44485",
+        "issue:44556",
+        "issue:44655",
+        "issue:44749",
+        "issue:44857",
+        "issue:44938",
+        "issue:44961",
+        "issue:44964",
+        "issue:44977",
+        "issue:45071",
+        "issue:45125",
+        "issue:45278",
+        "issue:45341",
+        "issue:45375",
+        "issue:45478",
+        "issue:45588",
+        "issue:45600",
+        "issue:45850"
+      ],
+      "soft_pairs": [
+        "issue:28282|issue:33453",
+        "issue:43618|issue:44857",
+        "issue:43618|issue:44977",
+        "issue:43618|issue:45588",
+        "issue:43901|issue:44938",
+        "issue:44655|issue:44938",
+        "issue:42890|issue:44964",
+        "issue:42890|issue:44112",
+        "issue:42890|issue:45600",
+        "issue:43502|issue:44062",
+        "issue:43010|issue:44075",
+        "issue:42890|issue:45341",
+        "issue:44222|issue:45125",
+        "issue:42548|issue:44351",
+        "issue:43502|issue:44485",
+        "issue:44355|issue:44749",
+        "issue:45125|issue:45478",
+        "issue:30990|issue:31795",
+        "issue:45125|issue:45375",
+        "issue:44387|issue:44556",
+        "issue:30333|issue:33453",
+        "issue:43010|issue:45278",
+        "issue:43723|issue:44556",
+        "issue:43502|issue:44222",
+        "issue:43010|issue:44368",
+        "issue:44315|issue:44360",
+        "issue:43010|issue:45850",
+        "issue:43502|issue:44961",
+        "issue:42831|issue:43874",
+        "issue:30333|issue:36296",
+        "issue:44355|issue:45071",
+        "issue:44355|issue:44655",
+        "issue:44222|issue:44230",
+        "issue:43827|issue:44961",
+        "issue:33453|issue:34689",
+        "issue:42757|issue:43576"
+      ],
+      "prepared_review_unit_hash": "d8f28d6806844c6ebf80142298da36d9466144be7a7039775e88452e7bc784a4",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 36,
+        "serialized_chars": 37954,
+        "estimated_input_tokens": 9489,
+        "estimated_eval_tokens": 19234
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "Mostly a grab-bag of unrelated issues across loading, modeling, tests, docs, and performance. None of the soft pairs are clearly the same underlying bug/change from titles alone.",
+        "confidence": 0.56,
+        "canonical_issue_reason": "Issue 45278 is the broadest umbrella for the visible v5 import-regression theme, so it is the most useful single issue to anchor this set.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "Issue 45278 is the best global tracker here because it is open, recent, and broad enough to subsume the generic import-failure cluster better than the narrower reports.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:28282",
+            "right": "issue:33453",
+            "accept": false,
+            "reason": "Different failures: PyTorch missing import vs tokenizer loading regression."
+          },
+          {
+            "left": "issue:43618",
+            "right": "issue:44857",
+            "accept": false,
+            "reason": "Unrelated model/output bug versus AMP/CUDA crash in a different loss path."
+          },
+          {
+            "left": "issue:43618",
+            "right": "issue:44977",
+            "accept": false,
+            "reason": "Different code paths: CLIP output attentions vs Qwen3.5 flash-attention generation."
+          },
+          {
+            "left": "issue:43618",
+            "right": "issue:45588",
+            "accept": false,
+            "reason": "Different failures in different attention code paths; not the same bug."
+          },
+          {
+            "left": "issue:43901",
+            "right": "issue:44938",
+            "accept": false,
+            "reason": "Docs mismatch for a pipeline argument vs Python 3.14 load failure."
+          },
+          {
+            "left": "issue:44655",
+            "right": "issue:44938",
+            "accept": false,
+            "reason": "Saving Pipeline objects is unrelated to Python 3.14 import/load issues."
+          },
+          {
+            "left": "issue:42890",
+            "right": "issue:44964",
+            "accept": false,
+            "reason": "Test seeding flake versus inability to load a specific multimodal model."
+          },
+          {
+            "left": "issue:42890",
+            "right": "issue:44112",
+            "accept": false,
+            "reason": "Both are test failures, but they describe different fixtures and root causes."
+          },
+          {
+            "left": "issue:42890",
+            "right": "issue:45600",
+            "accept": false,
+            "reason": "Different CI regressions: seed flakiness vs stale auto-mapping references."
+          },
+          {
+            "left": "issue:43502",
+            "right": "issue:44062",
+            "accept": false,
+            "reason": "Offline-mode request bug is unrelated to AddedToken keyword collision."
+          },
+          {
+            "left": "issue:43010",
+            "right": "issue:44075",
+            "accept": false,
+            "reason": "No-grad cache update bug is unrelated to optimizer argument handling."
+          },
+          {
+            "left": "issue:42890",
+            "right": "issue:45341",
+            "accept": false,
+            "reason": "Test seed flakiness is not the same as a helper bug in testing_utils.py."
+          },
+          {
+            "left": "issue:44222",
+            "right": "issue:45125",
+            "accept": false,
+            "reason": "FP8 save_pretrained issue and missing tensor-parallel plan are different model bugs."
+          },
+          {
+            "left": "issue:42548",
+            "right": "issue:44351",
+            "accept": false,
+            "reason": "Both are import errors, but for different symbols and likely different breakages."
+          },
+          {
+            "left": "issue:43502",
+            "right": "issue:44485",
+            "accept": false,
+            "reason": "Local-files-only network leak is unrelated to GLM-5 RoPE implementation."
+          },
+          {
+            "left": "issue:44355",
+            "right": "issue:44749",
+            "accept": false,
+            "reason": "Compiled-file errors and slowdown after upgrade are not the same defect."
+          },
+          {
+            "left": "issue:45125",
+            "right": "issue:45478",
+            "accept": false,
+            "reason": "Same model family, but tensor-parallel metadata and from_pretrained loading are distinct paths."
+          },
+          {
+            "left": "issue:30990",
+            "right": "issue:31795",
+            "accept": false,
+            "reason": "Both mention loading/docs, but one is Sentence Transformers hanging and the other is documentation confusion."
+          },
+          {
+            "left": "issue:45125",
+            "right": "issue:45375",
+            "accept": false,
+            "reason": "Missing _tp_plan and missing config field are different Qwen3.5 MoE issues."
+          },
+          {
+            "left": "issue:44387",
+            "right": "issue:44556",
+            "accept": false,
+            "reason": "Quantization memory regression is not the same as checkpoint reload incompatibility."
+          },
+          {
+            "left": "issue:30333",
+            "right": "issue:33453",
+            "accept": false,
+            "reason": "MLflow reporting on failed training jobs is unrelated to tokenizer loading regression."
+          },
+          {
+            "left": "issue:43010",
+            "right": "issue:45278",
+            "accept": false,
+            "reason": "Cache no_grad behavior is unrelated to broad import failures after upgrade."
+          },
+          {
+            "left": "issue:43723",
+            "right": "issue:44556",
+            "accept": false,
+            "reason": "Tokenizer loading and checkpoint reload problems overlap only loosely; likely different bugs."
+          },
+          {
+            "left": "issue:43502",
+            "right": "issue:44222",
+            "accept": false,
+            "reason": "Network-blocking bug and FP8 save_pretrained bug affect different code paths."
+          },
+          {
+            "left": "issue:43010",
+            "right": "issue:44368",
+            "accept": false,
+            "reason": "Cache update gradient handling is unrelated to a tied-embeddings warning."
+          },
+          {
+            "left": "issue:44315",
+            "right": "issue:44360",
+            "accept": false,
+            "reason": "Liger Kernel application bug and DSA indexer ReLU discussion are unrelated."
+          },
+          {
+            "left": "issue:43010",
+            "right": "issue:45850",
+            "accept": false,
+            "reason": "Specific cache/layer no_grad issue is not the same as a broad complaint about many model breakages."
+          },
+          {
+            "left": "issue:43502",
+            "right": "issue:44961",
+            "accept": false,
+            "reason": "Local-files-only networking bug is unrelated to the placeholder 'racoon' issue."
+          },
+          {
+            "left": "issue:42831",
+            "right": "issue:43874",
+            "accept": false,
+            "reason": "FP8 accuracy issue and missing multimodal token-count method are different defects."
+          },
+          {
+            "left": "issue:30333",
+            "right": "issue:36296",
+            "accept": false,
+            "reason": "MLflow reporting failure is unrelated to tensor-parallel training bugs."
+          },
+          {
+            "left": "issue:44355",
+            "right": "issue:45071",
+            "accept": false,
+            "reason": "Compiled-file load errors and PretrainedConfig type-checking regression are different issues."
+          },
+          {
+            "left": "issue:44355",
+            "right": "issue:44655",
+            "accept": false,
+            "reason": "Compiled-file execution errors and Pipeline save_pretrained failures are not the same bug."
+          },
+          {
+            "left": "issue:44222",
+            "right": "issue:44230",
+            "accept": false,
+            "reason": "Both involve FP8 and MoE, but one is saving and the other is inference support."
+          },
+          {
+            "left": "issue:43827",
+            "right": "issue:44961",
+            "accept": false,
+            "reason": "Docs still referencing pipeline() is unrelated to the unrelated 'racoon' issue."
+          },
+          {
+            "left": "issue:33453",
+            "right": "issue:34689",
+            "accept": false,
+            "reason": "Tokenizer loading regression and Llama 3.2 Vision model-loading breakage are different concrete bugs."
+          },
+          {
+            "left": "issue:42757",
+            "right": "issue:43576",
+            "accept": false,
+            "reason": "These are plausibly related import/runtime regressions, but the titles do not show the same concrete bug path strongly enough to merge."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 44,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:31795",
+        "issue:33453",
+        "issue:36296",
+        "issue:38617",
+        "issue:41628",
+        "issue:41669",
+        "issue:42548",
+        "issue:42890",
+        "issue:43010",
+        "issue:43299",
+        "issue:43352",
+        "issue:43408",
+        "issue:43525",
+        "issue:43526",
+        "issue:43572",
+        "issue:43576",
+        "issue:43673",
+        "issue:43824",
+        "issue:43901",
+        "issue:43976",
+        "issue:44038",
+        "issue:44246",
+        "issue:44261",
+        "issue:44273",
+        "issue:44295",
+        "issue:44297",
+        "issue:44315",
+        "issue:44351",
+        "issue:44355",
+        "issue:44360",
+        "issue:44373",
+        "issue:44393",
+        "issue:44556",
+        "issue:44623",
+        "issue:44704",
+        "issue:44908",
+        "issue:44961",
+        "issue:45003",
+        "issue:45071",
+        "issue:45092",
+        "issue:45125",
+        "issue:45230",
+        "issue:45362",
+        "issue:45468",
+        "issue:45542",
+        "issue:45656",
+        "issue:45753",
+        "issue:45850"
+      ],
+      "soft_pairs": [
+        "issue:43299|issue:44556",
+        "issue:43010|issue:45656",
+        "issue:33453|issue:36296",
+        "issue:43576|issue:45071",
+        "issue:43576|issue:43824",
+        "issue:44961|issue:45468",
+        "issue:31795|issue:38617",
+        "issue:38617|issue:43576",
+        "issue:41628|issue:42548",
+        "issue:43010|issue:43976",
+        "issue:43576|issue:44351",
+        "issue:44908|issue:44961",
+        "issue:36296|issue:43901",
+        "issue:41628|issue:44355",
+        "issue:36296|issue:43526",
+        "issue:44261|issue:44360",
+        "issue:44273|issue:45003",
+        "issue:44623|issue:45230",
+        "issue:44273|issue:45542",
+        "issue:44393|issue:45753",
+        "issue:43576|issue:44556",
+        "issue:44038|issue:45125",
+        "issue:44393|issue:45071",
+        "issue:43525|issue:45850",
+        "issue:41669|issue:43673",
+        "issue:44556|issue:45092",
+        "issue:44373|issue:44961",
+        "issue:43572|issue:45850",
+        "issue:45362|issue:45850",
+        "issue:44704|issue:44908",
+        "issue:42890|issue:44908",
+        "issue:44246|issue:44273",
+        "issue:43408|issue:44315",
+        "issue:43976|issue:44393",
+        "issue:43352|issue:44393",
+        "issue:43576|issue:44908",
+        "issue:44908|issue:45071",
+        "issue:44297|issue:45230",
+        "issue:44295|issue:45230",
+        "issue:41669|issue:44351"
+      ],
+      "prepared_review_unit_hash": "c4eb10c9fbcd48733eed069d34fb48f698788d37eb66002f5b5c298130e1000c",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 40,
+        "serialized_chars": 38941,
+        "estimated_input_tokens": 9736,
+        "estimated_eval_tokens": 19728
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "The cluster is very noisy: most pairs are only loosely related by version-5 regressions or import/lazy-loading symptoms, but they do not look like the same concrete bug or fix. I found no strong duplicate pair to merge, so the best anchor is a broad umbrella issue rather than a specific regression.",
+        "confidence": 0.38,
+        "canonical_issue_reason": "Issue 45850 is the broadest umbrella-style report in the set and best matches the noisy cluster of v5 breakage complaints, even though it is vague.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "Issue 45850 is the most suitable cluster representative because it is broad enough to cover multiple v5-related regressions without pretending the specific bugs are identical.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:43299",
+            "right": "issue:44556",
+            "accept": false,
+            "reason": "Both mention v5 loading breakage, but one is Qwen3VL MoE loading and the other is general checkpoint reload incompatibility; not the same concrete failure."
+          },
+          {
+            "left": "issue:43010",
+            "right": "issue:45656",
+            "accept": false,
+            "reason": "One is about cache/layer update no_grad decoration, the other about deepspeed calling optimizer.step twice; different code paths and symptoms."
+          },
+          {
+            "left": "issue:33453",
+            "right": "issue:36296",
+            "accept": false,
+            "reason": "Tokenizer loading regression and tensor-parallel training bug are unrelated subsystems and failures."
+          },
+          {
+            "left": "issue:43576",
+            "right": "issue:45071",
+            "accept": false,
+            "reason": "Broken env command and PretrainedConfig type-checking regression are separate issues with no shared code path."
+          },
+          {
+            "left": "issue:43576",
+            "right": "issue:43824",
+            "accept": false,
+            "reason": "Env command failure is unrelated to a missing Qwen2.5-VL top-level import."
+          },
+          {
+            "left": "issue:44961",
+            "right": "issue:45468",
+            "accept": false,
+            "reason": "The titles are unrelated and provide no evidence of the same bug or change."
+          },
+          {
+            "left": "issue:31795",
+            "right": "issue:38617",
+            "accept": false,
+            "reason": "Documentation confusion for forward args is unrelated to an ImportError for layer_type_validation."
+          },
+          {
+            "left": "issue:38617",
+            "right": "issue:43576",
+            "accept": false,
+            "reason": "An import error for a specific symbol and a broken CLI env command are different problems."
+          },
+          {
+            "left": "issue:41628",
+            "right": "issue:42548",
+            "accept": false,
+            "reason": "Both are import errors, but for different top-level classes; the titles do not indicate the same missing-export bug."
+          },
+          {
+            "left": "issue:43010",
+            "right": "issue:43976",
+            "accept": false,
+            "reason": "A no_grad/decorator request and a Python-version compatibility issue are not the same defect."
+          },
+          {
+            "left": "issue:43576",
+            "right": "issue:44351",
+            "accept": false,
+            "reason": "Broken env command and missing HybridCache import are unrelated."
+          },
+          {
+            "left": "issue:44908",
+            "right": "issue:44961",
+            "accept": false,
+            "reason": "Scheduler kwargs handling and a nonsensical titled issue are unrelated."
+          },
+          {
+            "left": "issue:36296",
+            "right": "issue:43901",
+            "accept": false,
+            "reason": "Tensor-parallel training and text-classification docs behavior are not the same underlying problem."
+          },
+          {
+            "left": "issue:41628",
+            "right": "issue:44355",
+            "accept": false,
+            "reason": "A missing AutoImageProcessor import and errors running compiled Python files do not point to the same bug."
+          },
+          {
+            "left": "issue:36296",
+            "right": "issue:43526",
+            "accept": false,
+            "reason": "Tensor parallel training bug and BeitImageProcessorFast reduce_labels bug are unrelated."
+          },
+          {
+            "left": "issue:44261",
+            "right": "issue:44360",
+            "accept": false,
+            "reason": "MLA q_a_layernorm precision/config issue and missing ReLU in the DSA indexer are different model changes."
+          },
+          {
+            "left": "issue:44273",
+            "right": "issue:45003",
+            "accept": false,
+            "reason": "Lazy loading issues and unsafe sys.modules access may be adjacent, but the titles do not establish the same concrete bug."
+          },
+          {
+            "left": "issue:44623",
+            "right": "issue:45230",
+            "accept": false,
+            "reason": "Processor.save_pretrained missing files is unrelated to a generic bug-report placeholder."
+          },
+          {
+            "left": "issue:44273",
+            "right": "issue:45542",
+            "accept": false,
+            "reason": "Lazy loading and a tf backend error from tensorboard-only installs are different problems."
+          },
+          {
+            "left": "issue:44393",
+            "right": "issue:45753",
+            "accept": false,
+            "reason": "Both are Qwen3-related, but one is bounding-box hallucination and the other is an infinite loop on a specific image; too different to merge."
+          },
+          {
+            "left": "issue:43576",
+            "right": "issue:44556",
+            "accept": false,
+            "reason": "Broken env command and old checkpoint reload incompatibility are separate regressions."
+          },
+          {
+            "left": "issue:44038",
+            "right": "issue:45125",
+            "accept": false,
+            "reason": "Both touch Qwen3 MoE, but one is a general v5/Qwen3-VL-Moe issue and the other is a missing tensor-parallel plan; not the same code-path problem."
+          },
+          {
+            "left": "issue:44393",
+            "right": "issue:45071",
+            "accept": false,
+            "reason": "Qwen3-VL output bugs and PretrainedConfig type-checking are unrelated."
+          },
+          {
+            "left": "issue:43525",
+            "right": "issue:45850",
+            "accept": false,
+            "reason": "A missing pad_token_id on Llama4Config is one specific model config bug, not the broad generic complaint in 45850."
+          },
+          {
+            "left": "issue:41669",
+            "right": "issue:43673",
+            "accept": false,
+            "reason": "Removing import-* usage for import-time performance and a missing GenerationMixin cache during chunked prefill are unrelated."
+          },
+          {
+            "left": "issue:44556",
+            "right": "issue:45092",
+            "accept": false,
+            "reason": "Both are about v5 incompatibility, but one is general checkpoint reload after upgrade and the other is a specific remote-code InternVL2/meta-init issue; not clearly the same bug."
+          },
+          {
+            "left": "issue:44373",
+            "right": "issue:44961",
+            "accept": false,
+            "reason": "Wrong docstring for position_ids is unrelated to the placeholder racoon issue."
+          },
+          {
+            "left": "issue:43572",
+            "right": "issue:45850",
+            "accept": false,
+            "reason": "StableLmConfig missing pad_token_idx and the broad 'v5 breaks too many models' complaint are not the same concrete regression."
+          },
+          {
+            "left": "issue:45362",
+            "right": "issue:45850",
+            "accept": false,
+            "reason": "Qwen3.5 chat crashes and a generic v5 breakage complaint are too broad to call duplicates."
+          },
+          {
+            "left": "issue:44704",
+            "right": "issue:44908",
+            "accept": false,
+            "reason": "AutoProcessor kwargs forwarding and inverse_sqrt scheduler kwargs are completely different code paths."
+          },
+          {
+            "left": "issue:42890",
+            "right": "issue:44908",
+            "accept": false,
+            "reason": "A flaky integration test and a scheduler kwargs bug are unrelated."
+          },
+          {
+            "left": "issue:44246",
+            "right": "issue:44273",
+            "accept": false,
+            "reason": "Slow import and lazy-loading malfunction are related only at a very high level; the titles do not show the same underlying defect."
+          },
+          {
+            "left": "issue:43408",
+            "right": "issue:44315",
+            "accept": false,
+            "reason": "sam3 tracker/model-type warning and Liger Kernel not applied with model_init are unrelated."
+          },
+          {
+            "left": "issue:43976",
+            "right": "issue:44393",
+            "accept": false,
+            "reason": "Python-version support issue and Qwen3-VL bounding-box output bugs are unrelated."
+          },
+          {
+            "left": "issue:43352",
+            "right": "issue:44393",
+            "accept": false,
+            "reason": "Flash Attention support in Nemotron and Qwen3-VL hallucination/output errors are different issues."
+          },
+          {
+            "left": "issue:43576",
+            "right": "issue:44908",
+            "accept": false,
+            "reason": "A broken env command and scheduler kwargs handling do not share the same bug."
+          },
+          {
+            "left": "issue:44297",
+            "right": "issue:45230",
+            "accept": false,
+            "reason": "Tokenizer save_pretrained file mismatch and a generic bug report are not the same issue."
+          },
+          {
+            "left": "issue:44295",
+            "right": "issue:45230",
+            "accept": false,
+            "reason": "Reading registered position_ids as a buffer and a generic bug report are unrelated."
+          },
+          {
+            "left": "issue:41669",
+            "right": "issue:44351",
+            "accept": false,
+            "reason": "Import-* performance cleanup and HybridCache import failure are different problems."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 2,
+      "component_count": 16,
+      "review_unit_index": 45,
+      "review_unit_count": 45,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-15354-392",
+      "nodes": [
+        "issue:30333",
+        "issue:31515",
+        "issue:31795",
+        "issue:33453",
+        "issue:34689",
+        "issue:36296",
+        "issue:37428",
+        "issue:38617",
+        "issue:41669",
+        "issue:42371",
+        "issue:42913",
+        "issue:43066",
+        "issue:43298",
+        "issue:43502",
+        "issue:43519",
+        "issue:43525",
+        "issue:43572",
+        "issue:43576",
+        "issue:43606",
+        "issue:43673",
+        "issue:43704",
+        "issue:43827",
+        "issue:43874",
+        "issue:44077",
+        "issue:44230",
+        "issue:44261",
+        "issue:44263",
+        "issue:44273",
+        "issue:44315",
+        "issue:44360",
+        "issue:44393",
+        "issue:44485",
+        "issue:44492",
+        "issue:44623",
+        "issue:44655",
+        "issue:44829",
+        "issue:44861",
+        "issue:44908",
+        "issue:44998",
+        "issue:45092",
+        "issue:45230",
+        "issue:45278",
+        "issue:45290",
+        "issue:45310",
+        "issue:45341",
+        "issue:45468",
+        "issue:45478",
+        "issue:45850"
+      ],
+      "soft_pairs": [
+        "issue:43673|issue:44492",
+        "issue:41669|issue:45341",
+        "issue:45278|issue:45850",
+        "issue:44393|issue:45478",
+        "issue:43874|issue:45850",
+        "issue:41669|issue:44492",
+        "issue:44315|issue:44829",
+        "issue:43827|issue:44908",
+        "issue:44077|issue:44861",
+        "issue:31795|issue:33453",
+        "issue:41669|issue:44908",
+        "issue:42371|issue:44492",
+        "issue:43576|issue:44273",
+        "issue:44655|issue:44861",
+        "issue:44360|issue:45468",
+        "issue:44492|issue:45850",
+        "issue:44273|issue:44908",
+        "issue:45230|issue:45310",
+        "issue:43519|issue:44485",
+        "issue:41669|issue:43704",
+        "issue:43519|issue:44908",
+        "issue:31795|issue:34689",
+        "issue:31795|issue:37428",
+        "issue:44315|issue:45092",
+        "issue:43704|issue:44485",
+        "issue:44485|issue:45468",
+        "issue:42371|issue:43704",
+        "issue:43502|issue:43519",
+        "issue:43519|issue:44623",
+        "issue:43519|issue:43704",
+        "issue:44263|issue:45468",
+        "issue:43606|issue:45468",
+        "issue:44261|issue:45468",
+        "issue:43704|issue:45341",
+        "issue:42913|issue:43066",
+        "issue:44908|issue:44998",
+        "issue:44861|issue:45290",
+        "issue:43704|issue:44998",
+        "issue:41669|issue:44998",
+        "issue:43525|issue:44861",
+        "issue:43066|issue:44998",
+        "issue:43572|issue:44861",
+        "issue:31515|issue:38617",
+        "issue:44360|issue:44998",
+        "issue:44230|issue:44623",
+        "issue:30333|issue:31515",
+        "issue:31515|issue:36296",
+        "issue:31515|issue:31795",
+        "issue:44230|issue:45310",
+        "issue:31515|issue:33453",
+        "issue:43298|issue:44230"
+      ],
+      "prepared_review_unit_hash": "77388cf968112299073a9ee4618f3332fdd907d3d6722246b268625fcb62e74e",
+      "component_budget": {
+        "node_count": 392,
+        "item_count": 392,
+        "soft_pair_count": 1480,
+        "serialized_chars": 761482,
+        "estimated_input_tokens": 190371,
+        "estimated_eval_tokens": 380998
+      },
+      "budget": {
+        "node_count": 48,
+        "item_count": 48,
+        "soft_pair_count": 51,
+        "serialized_chars": 42267,
+        "estimated_input_tokens": 10567,
+        "estimated_eval_tokens": 21390
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "Mostly unrelated Transformer issues were grouped together. Several are broad v5 regression/loading complaints, but the pairwise evidence is too weak to treat any soft edge as a true duplicate.",
+        "confidence": 0.83,
+        "canonical_issue_reason": "issue:45278 is the best representative of the broad v5 regression/load-failure theme: it is specific enough to be actionable, but still general enough to cover the nearby model-loading complaints in this cluster.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "issue:45278 is the most central and reusable issue here; the others are either narrower sub-bugs, documentation issues, or too vague to serve as a stable canonical.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:43673",
+            "right": "issue:44492",
+            "accept": false,
+            "reason": "Cache-related titles, but one is a generation cache bug and the other is a typo/documentation issue."
+          },
+          {
+            "left": "issue:41669",
+            "right": "issue:45341",
+            "accept": false,
+            "reason": "Unrelated subjects: import-star cleanup versus a testing_utils bug."
+          },
+          {
+            "left": "issue:45278",
+            "right": "issue:45850",
+            "accept": false,
+            "reason": "Both are broad v5 complaints, but not the same concrete bug or code path."
+          },
+          {
+            "left": "issue:44393",
+            "right": "issue:45478",
+            "accept": false,
+            "reason": "Different models and failures: Qwen3-VL bbox output versus Qwen3.5 MoE from_pretrained loading."
+          },
+          {
+            "left": "issue:43874",
+            "right": "issue:45850",
+            "accept": false,
+            "reason": "Specific GLM46V image-processor AttributeError versus a generic many-models breakage report."
+          },
+          {
+            "left": "issue:41669",
+            "right": "issue:44492",
+            "accept": false,
+            "reason": "Cleanup/import-performance issue versus cache-strategy typo; not the same bug."
+          },
+          {
+            "left": "issue:44315",
+            "right": "issue:44829",
+            "accept": false,
+            "reason": "Both mention training, but the code paths are different: model_init/Liger Kernel versus flash_attention_3 degenerate training."
+          },
+          {
+            "left": "issue:43827",
+            "right": "issue:44908",
+            "accept": false,
+            "reason": "Documentation after pipeline removals versus a scheduler kwargs bug."
+          },
+          {
+            "left": "issue:44077",
+            "right": "issue:44861",
+            "accept": false,
+            "reason": "Different failures in different utilities; no shared underlying bug."
+          },
+          {
+            "left": "issue:31795",
+            "right": "issue:33453",
+            "accept": false,
+            "reason": "Documentation confusion versus tokenizer loading regression."
+          },
+          {
+            "left": "issue:41669",
+            "right": "issue:44908",
+            "accept": false,
+            "reason": "Import-star cleanup and inverse_sqrt scheduler kwargs are unrelated."
+          },
+          {
+            "left": "issue:42371",
+            "right": "issue:44492",
+            "accept": false,
+            "reason": "TF32 API migration versus a cache-strategy typo."
+          },
+          {
+            "left": "issue:43576",
+            "right": "issue:44273",
+            "accept": false,
+            "reason": "Broken env command and lazy-loading malfunction are separate issues."
+          },
+          {
+            "left": "issue:44655",
+            "right": "issue:44861",
+            "accept": false,
+            "reason": "save_pretrained pipeline export issue versus tied-weights-key crash."
+          },
+          {
+            "left": "issue:44360",
+            "right": "issue:45468",
+            "accept": false,
+            "reason": "Different model/component bugs: DSA indexer ReLU versus Gemma4Audio positional encoding."
+          },
+          {
+            "left": "issue:44492",
+            "right": "issue:45850",
+            "accept": false,
+            "reason": "A typo issue does not match a generic v5 breakage report."
+          },
+          {
+            "left": "issue:44273",
+            "right": "issue:44908",
+            "accept": false,
+            "reason": "Lazy loading and scheduler kwargs are unrelated subsystems."
+          },
+          {
+            "left": "issue:45230",
+            "right": "issue:45310",
+            "accept": false,
+            "reason": "Generic bug report versus a specific Qwen3.5 MoE from_pretrained error; not enough to merge."
+          },
+          {
+            "left": "issue:43519",
+            "right": "issue:44485",
+            "accept": false,
+            "reason": "Timestamp calculation in Qwen3VL is unrelated to GLM-5 RoPE implementation."
+          },
+          {
+            "left": "issue:41669",
+            "right": "issue:43704",
+            "accept": false,
+            "reason": "Import cleanup versus VRAM leak; no common code-path."
+          },
+          {
+            "left": "issue:43519",
+            "right": "issue:44908",
+            "accept": false,
+            "reason": "Different components: processor timestamp math versus scheduler behavior."
+          },
+          {
+            "left": "issue:31795",
+            "right": "issue:34689",
+            "accept": false,
+            "reason": "Documentation ambiguity versus a model-loading regression."
+          },
+          {
+            "left": "issue:31795",
+            "right": "issue:37428",
+            "accept": false,
+            "reason": "Doc issue versus import error from flash attention utils."
+          },
+          {
+            "left": "issue:44315",
+            "right": "issue:45092",
+            "accept": false,
+            "reason": "Both touch model initialization, but one is about Liger Kernel application and the other about remote-code/meta init incompatibility."
+          },
+          {
+            "left": "issue:43704",
+            "right": "issue:44485",
+            "accept": false,
+            "reason": "VRAM leak and RoPE implementation are unrelated bugs."
+          },
+          {
+            "left": "issue:44485",
+            "right": "issue:45468",
+            "accept": false,
+            "reason": "Different model implementations and failure modes."
+          },
+          {
+            "left": "issue:42371",
+            "right": "issue:43704",
+            "accept": false,
+            "reason": "TF32 API guidance and VRAM leakage do not share a code path."
+          },
+          {
+            "left": "issue:43502",
+            "right": "issue:43519",
+            "accept": false,
+            "reason": "Local-files-only network leak versus tokenizer/processor timestamp bug."
+          },
+          {
+            "left": "issue:43519",
+            "right": "issue:44623",
+            "accept": false,
+            "reason": "Timestamp calculation bug versus missing files on save_pretrained."
+          },
+          {
+            "left": "issue:43519",
+            "right": "issue:43704",
+            "accept": false,
+            "reason": "Processor timestamp bug versus multi-thread VRAM leak."
+          },
+          {
+            "left": "issue:44263",
+            "right": "issue:45468",
+            "accept": false,
+            "reason": "torch.split return values in a GLM indexer and Gemma audio positional encoding are different issues."
+          },
+          {
+            "left": "issue:43606",
+            "right": "issue:45468",
+            "accept": false,
+            "reason": "CPU offload device mismatch is unrelated to Gemma audio positional encoding."
+          },
+          {
+            "left": "issue:44261",
+            "right": "issue:45468",
+            "accept": false,
+            "reason": "MLA layernorm epsilon precision issue versus a Gemma audio encoding bug."
+          },
+          {
+            "left": "issue:43704",
+            "right": "issue:45341",
+            "accept": false,
+            "reason": "VRAM leak and a testing_utils bug are unrelated."
+          },
+          {
+            "left": "issue:42913",
+            "right": "issue:43066",
+            "accept": false,
+            "reason": "Both are tokenizer-v5 complaints, but one is a broad behavior regression and the other is a specific decoder-type error; too different to merge."
+          },
+          {
+            "left": "issue:44908",
+            "right": "issue:44998",
+            "accept": false,
+            "reason": "Scheduler bug versus an unrelated placeholder issue."
+          },
+          {
+            "left": "issue:44861",
+            "right": "issue:45290",
+            "accept": false,
+            "reason": "Tied-weights key handling and chat-template tool-call crashing are separate code paths."
+          },
+          {
+            "left": "issue:43704",
+            "right": "issue:44998",
+            "accept": false,
+            "reason": "VRAM leak report versus unrelated placeholder issue."
+          },
+          {
+            "left": "issue:44230",
+            "right": "issue:45310",
+            "accept": false,
+            "reason": "Feature/support request for fp8 infer support versus a from_pretrained loading error."
+          },
+          {
+            "left": "issue:31515",
+            "right": "issue:38617",
+            "accept": false,
+            "reason": "Slow checkpoint loading and an import error are unrelated."
+          },
+          {
+            "left": "issue:44360",
+            "right": "issue:44998",
+            "accept": false,
+            "reason": "DSA indexer bug versus unrelated placeholder issue."
+          },
+          {
+            "left": "issue:44230",
+            "right": "issue:44623",
+            "accept": false,
+            "reason": "FP8 support request versus processor save_pretrained missing files."
+          },
+          {
+            "left": "issue:30333",
+            "right": "issue:31515",
+            "accept": false,
+            "reason": "MLFlow training-job reporting and checkpoint-loading speed are different problems."
+          },
+          {
+            "left": "issue:31515",
+            "right": "issue:36296",
+            "accept": false,
+            "reason": "Checkpoint loading slowdown versus tensor-parallel training bug."
+          },
+          {
+            "left": "issue:31515",
+            "right": "issue:31795",
+            "accept": false,
+            "reason": "Slow loading versus documentation confusion; not the same issue."
+          },
+          {
+            "left": "issue:44230",
+            "right": "issue:45310",
+            "accept": false,
+            "reason": "FP8 support request and Qwen3.5 MoE from_pretrained error are different underlying changes."
+          },
+          {
+            "left": "issue:31515",
+            "right": "issue:33453",
+            "accept": false,
+            "reason": "Loading speed regression versus tokenizer loading regression."
+          },
+          {
+            "left": "issue:43298",
+            "right": "issue:44230",
+            "accept": false,
+            "reason": "Continuous batching audio-model support and fp8 inference support are unrelated features."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 3,
+      "component_count": 16,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-26350-2",
+      "nodes": [
+        "issue:26350",
+        "issue:33260"
+      ],
+      "soft_pairs": [
+        "issue:26350|issue:33260"
+      ],
+      "prepared_review_unit_hash": "07cce9baa57ceac172565ceae10b960646de03b3e915759bd7f47d56b17f6535",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 1589,
+        "estimated_input_tokens": 398,
+        "estimated_eval_tokens": 1052
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 1589,
+        "estimated_input_tokens": 398,
+        "estimated_eval_tokens": 1052
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "These are two unrelated community feature requests: one for Flash Attention 2 architecture support and one for GGUF support across more architectures. They share only a broad \u201cexpand architecture support\u201d theme, not the same underlying change.",
+        "confidence": 0.98,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": null,
+        "best_issue_reason": "Issue #26350 is the older and more discussed request, so it is the better anchor if one issue must represent the cluster; however, it is not a duplicate of #33260.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:26350",
+            "right": "issue:33260",
+            "accept": false,
+            "reason": "Different feature requests for different subsystems: Flash Attention 2 support vs GGUF support. They do not describe the same bug or the same code change, so they should not be merged as duplicates."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 4,
+      "component_count": 16,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-27088-4",
+      "nodes": [
+        "issue:27088",
+        "issue:43834",
+        "issue:44365",
+        "issue:44370"
+      ],
+      "soft_pairs": [
+        "issue:27088|issue:44370",
+        "issue:27088|issue:44365",
+        "issue:27088|issue:43834",
+        "issue:43834|issue:44370",
+        "issue:43834|issue:44365",
+        "issue:44365|issue:44370"
+      ],
+      "prepared_review_unit_hash": "da07ef3e1f3e2a5f6d0e74ea80aa590d5d2811aaac92dd5b417fa6b9fc839e84",
+      "component_budget": {
+        "node_count": 4,
+        "item_count": 4,
+        "soft_pair_count": 6,
+        "serialized_chars": 4454,
+        "estimated_input_tokens": 1114,
+        "estimated_eval_tokens": 2484
+      },
+      "budget": {
+        "node_count": 4,
+        "item_count": 4,
+        "soft_pair_count": 6,
+        "serialized_chars": 4454,
+        "estimated_input_tokens": 1114,
+        "estimated_eval_tokens": 2484
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "These are issue-template variants for translating docs into different languages; only one is a concrete Turkish instance, while the others are generic placeholders. They are similar in wording but not the same underlying issue.",
+        "confidence": 0.93,
+        "canonical_issue_reason": "issue:27088 is the only concrete, language-specific report ([i18n-TR] Turkish), so it is the most informative representative issue.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "issue:27088 is the best global issue because it is specific and actionable, whereas the others are generic template entries without a distinct concrete language target.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:27088",
+            "right": "issue:44370",
+            "accept": false,
+            "reason": "Different issue instances: Turkish translation vs. a generic placeholder-language template; not the same concrete task."
+          },
+          {
+            "left": "issue:27088",
+            "right": "issue:44365",
+            "accept": false,
+            "reason": "Different issue instances: Turkish translation vs. a generic placeholder-language template; not the same concrete task."
+          },
+          {
+            "left": "issue:27088",
+            "right": "issue:43834",
+            "accept": false,
+            "reason": "Different issue instances: Turkish translation vs. a generic placeholder-language template; not the same concrete task."
+          },
+          {
+            "left": "issue:43834",
+            "right": "issue:44370",
+            "accept": false,
+            "reason": "Both are generic template-style issues with no concrete shared language target shown; too ambiguous to treat as the same underlying issue."
+          },
+          {
+            "left": "issue:43834",
+            "right": "issue:44365",
+            "accept": false,
+            "reason": "Both are generic template-style issues with no concrete shared language target shown; too ambiguous to treat as the same underlying issue."
+          },
+          {
+            "left": "issue:44365",
+            "right": "issue:44370",
+            "accept": false,
+            "reason": "Both are generic template-style issues with no concrete shared language target shown; too ambiguous to treat as the same underlying issue."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 5,
+      "component_count": 16,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-36817-15",
+      "nodes": [
+        "issue:36817",
+        "issue:36879",
+        "issue:37015",
+        "issue:41115",
+        "issue:41211",
+        "issue:42503",
+        "issue:42738",
+        "issue:43671",
+        "issue:43837",
+        "issue:43909",
+        "issue:44121",
+        "issue:44205",
+        "issue:45027",
+        "issue:45177",
+        "issue:45322"
+      ],
+      "soft_pairs": [
+        "issue:36817|issue:41211",
+        "issue:36817|issue:43671",
+        "issue:36817|issue:36879",
+        "issue:36879|issue:41211",
+        "issue:36879|issue:43909",
+        "issue:36879|issue:44121",
+        "issue:36879|issue:43837",
+        "issue:36879|issue:43671",
+        "issue:37015|issue:43837",
+        "issue:37015|issue:44121",
+        "issue:37015|issue:43909",
+        "issue:37015|issue:42503",
+        "issue:36879|issue:37015",
+        "issue:41115|issue:43837",
+        "issue:41115|issue:44121",
+        "issue:41115|issue:42503",
+        "issue:41115|issue:45027",
+        "issue:41211|issue:43671",
+        "issue:42503|issue:43837",
+        "issue:42503|issue:43671",
+        "issue:42503|issue:42738",
+        "issue:42738|issue:45027",
+        "issue:43671|issue:43837",
+        "issue:43671|issue:45027",
+        "issue:43671|issue:45177",
+        "issue:43671|issue:44121",
+        "issue:43837|issue:45322",
+        "issue:43837|issue:43909",
+        "issue:43909|issue:44205",
+        "issue:43671|issue:43909",
+        "issue:44121|issue:45322",
+        "issue:43837|issue:44121",
+        "issue:43837|issue:45027",
+        "issue:37015|issue:45027",
+        "issue:43837|issue:45177"
+      ],
+      "prepared_review_unit_hash": "425ebecca23ffca7bea528964385504261a9688191e42e55982b42baa67b5521",
+      "component_budget": {
+        "node_count": 15,
+        "item_count": 15,
+        "soft_pair_count": 35,
+        "serialized_chars": 20700,
+        "estimated_input_tokens": 5175,
+        "estimated_eval_tokens": 10606
+      },
+      "budget": {
+        "node_count": 15,
+        "item_count": 15,
+        "soft_pair_count": 35,
+        "serialized_chars": 20700,
+        "estimated_input_tokens": 5175,
+        "estimated_eval_tokens": 10606
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "These are not duplicates; they are separate model-request issues spanning different models and modalities (BERT variants, detection, audio/TTS/ASR, OCR, vision, and a sparse transformer).",
+        "confidence": 0.96,
+        "canonical_issue_reason": "No single canonical duplicate exists here; the cluster is a set of distinct feature requests. If one must be chosen as the broadest representative, issue 42738 is the closest umbrella item.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "Issue 42738 ('BERT-like models with RoPE') is the broadest and most reusable representative, but it still only fits part of the cluster and does not cover the audio/vision-specific requests.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:36817",
+            "right": "issue:41211",
+            "accept": false,
+            "reason": "Different requested models (EuroBert vs DEIMv2); same general area of model support, but not the same change."
+          },
+          {
+            "left": "issue:36817",
+            "right": "issue:43671",
+            "accept": false,
+            "reason": "EuroBert config support is unrelated to Qwen3-TTS support; different modality and code path."
+          },
+          {
+            "left": "issue:36817",
+            "right": "issue:36879",
+            "accept": false,
+            "reason": "EuroBert and RF-DETR are different models with different integration needs."
+          },
+          {
+            "left": "issue:36879",
+            "right": "issue:41211",
+            "accept": false,
+            "reason": "RF-DETR detection support and DEIMv2 are separate model additions."
+          },
+          {
+            "left": "issue:36879",
+            "right": "issue:43909",
+            "accept": false,
+            "reason": "RF-DETR is a detection model; LFM2.5 Audio is an audio model request."
+          },
+          {
+            "left": "issue:36879",
+            "right": "issue:44121",
+            "accept": false,
+            "reason": "Different model families and implementation targets; not the same bug/change."
+          },
+          {
+            "left": "issue:36879",
+            "right": "issue:43837",
+            "accept": false,
+            "reason": "RF-DETR and Qwen3-ASR are different model requests with different runtime paths."
+          },
+          {
+            "left": "issue:36879",
+            "right": "issue:43671",
+            "accept": false,
+            "reason": "RF-DETR vs Qwen3-TTS are distinct model integrations, not duplicates."
+          },
+          {
+            "left": "issue:37015",
+            "right": "issue:43837",
+            "accept": false,
+            "reason": "NeoBERT and Qwen3-ASR are different model additions and would not merge into one PR."
+          },
+          {
+            "left": "issue:37015",
+            "right": "issue:44121",
+            "accept": false,
+            "reason": "NeoBERT and the sparse transformer request are unrelated model architectures."
+          },
+          {
+            "left": "issue:37015",
+            "right": "issue:43909",
+            "accept": false,
+            "reason": "NeoBERT (text model) and LFM2.5 Audio are different modalities and code paths."
+          },
+          {
+            "left": "issue:37015",
+            "right": "issue:42503",
+            "accept": false,
+            "reason": "NeoBERT and ModernVBERT are separate model requests, not the same underlying change."
+          },
+          {
+            "left": "issue:36879",
+            "right": "issue:37015",
+            "accept": false,
+            "reason": "RF-DETR and NeoBERT are unrelated model support requests."
+          },
+          {
+            "left": "issue:41115",
+            "right": "issue:43837",
+            "accept": false,
+            "reason": "MiniCPM3 architecture support and Qwen3-ASR support are different model integrations."
+          },
+          {
+            "left": "issue:41115",
+            "right": "issue:44121",
+            "accept": false,
+            "reason": "MiniCPM3 architecture and the sparse transformer request are distinct code additions."
+          },
+          {
+            "left": "issue:41115",
+            "right": "issue:42503",
+            "accept": false,
+            "reason": "MiniCPM3 and ModernVBERT are separate model families."
+          },
+          {
+            "left": "issue:41115",
+            "right": "issue:45027",
+            "accept": false,
+            "reason": "MiniCPM3 architecture support and Voxtral TTS support are different modalities and code paths."
+          },
+          {
+            "left": "issue:41211",
+            "right": "issue:43671",
+            "accept": false,
+            "reason": "DEIMv2 and Qwen3-TTS are different model requests with no shared concrete fix."
+          },
+          {
+            "left": "issue:42503",
+            "right": "issue:43837",
+            "accept": false,
+            "reason": "ModernVBERT models and Qwen3-ASR are different model support requests."
+          },
+          {
+            "left": "issue:42503",
+            "right": "issue:43671",
+            "accept": false,
+            "reason": "ModernVBERT and Qwen3-TTS are distinct model additions."
+          },
+          {
+            "left": "issue:42503",
+            "right": "issue:42738",
+            "accept": false,
+            "reason": "ModernVBERT models are specific additions; 'BERT-like models with RoPE' is only a broad category, not the same request."
+          },
+          {
+            "left": "issue:42738",
+            "right": "issue:45027",
+            "accept": false,
+            "reason": "BERT-like RoPE models and Voxtral TTS are different model families and implementation paths."
+          },
+          {
+            "left": "issue:43671",
+            "right": "issue:43837",
+            "accept": false,
+            "reason": "Qwen3-TTS and Qwen3-ASR are related by naming only; they are separate tasks and code paths."
+          },
+          {
+            "left": "issue:43671",
+            "right": "issue:45027",
+            "accept": false,
+            "reason": "Qwen3-TTS and Voxtral TTS are different concrete model-support requests."
+          },
+          {
+            "left": "issue:43671",
+            "right": "issue:45177",
+            "accept": false,
+            "reason": "TTS support and DeepSeek-OCR2 are unrelated modalities."
+          },
+          {
+            "left": "issue:43671",
+            "right": "issue:44121",
+            "accept": false,
+            "reason": "Qwen3-TTS and the sparse transformer request are different underlying changes."
+          },
+          {
+            "left": "issue:43837",
+            "right": "issue:45322",
+            "accept": false,
+            "reason": "Qwen3-ASR support and EUPE encoder support are different model integrations."
+          },
+          {
+            "left": "issue:43837",
+            "right": "issue:43909",
+            "accept": false,
+            "reason": "ASR and audio generation model support are distinct requests, not one bug/change."
+          },
+          {
+            "left": "issue:43909",
+            "right": "issue:44205",
+            "accept": false,
+            "reason": "LFM2.5 Audio and SAM3-LiteText are different model families and modalities."
+          },
+          {
+            "left": "issue:43671",
+            "right": "issue:43909",
+            "accept": false,
+            "reason": "Qwen3-TTS and LFM2.5 Audio are separate audio-model requests."
+          },
+          {
+            "left": "issue:44121",
+            "right": "issue:45322",
+            "accept": false,
+            "reason": "Sparse transformer support and EUPE support are different architectures."
+          },
+          {
+            "left": "issue:43837",
+            "right": "issue:44121",
+            "accept": false,
+            "reason": "Qwen3-ASR and the sparse transformer request do not target the same code path."
+          },
+          {
+            "left": "issue:43837",
+            "right": "issue:45027",
+            "accept": false,
+            "reason": "Qwen3-ASR and Voxtral TTS are different speech-model additions."
+          },
+          {
+            "left": "issue:37015",
+            "right": "issue:45027",
+            "accept": false,
+            "reason": "NeoBERT and Voxtral TTS are unrelated model support requests."
+          },
+          {
+            "left": "issue:43837",
+            "right": "issue:45177",
+            "accept": false,
+            "reason": "ASR support and OCR2 support are different modalities and not duplicates."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 6,
+      "component_count": 16,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-37934-6",
+      "nodes": [
+        "issue:37934",
+        "issue:42936",
+        "issue:43240",
+        "issue:43668",
+        "issue:43813",
+        "issue:44031"
+      ],
+      "soft_pairs": [
+        "issue:37934|issue:43668",
+        "issue:42936|issue:43240",
+        "issue:42936|issue:44031",
+        "issue:42936|issue:43813",
+        "issue:42936|issue:43668",
+        "issue:43240|issue:43668",
+        "issue:43668|issue:44031",
+        "issue:43668|issue:43813"
+      ],
+      "prepared_review_unit_hash": "7d746c0b3b1d7c314490fa1a3a5666b7717f8c809480ad90a20bf5b8ea1285d9",
+      "component_budget": {
+        "node_count": 6,
+        "item_count": 6,
+        "soft_pair_count": 8,
+        "serialized_chars": 6293,
+        "estimated_input_tokens": 1574,
+        "estimated_eval_tokens": 3404
+      },
+      "budget": {
+        "node_count": 6,
+        "item_count": 6,
+        "soft_pair_count": 8,
+        "serialized_chars": 6293,
+        "estimated_input_tokens": 1574,
+        "estimated_eval_tokens": 3404
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "The cluster is not a true duplicate set: the issues concern unrelated areas (normalization, Mask2Former loss handling, tokenizer warnings, a config type hint, and a typo in a quantization symbol). The only commonality is superficial wording, not the same underlying bug.",
+        "confidence": 0.91,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": null,
+        "best_issue_reason": null,
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:37934",
+            "right": "issue:43668",
+            "accept": false,
+            "reason": "Both mention normalization-related terms, but one is about whether Llama4TextL2Norm is RMS norm and the other is a ModernBERTConfig type hint; different files, models, and bug types."
+          },
+          {
+            "left": "issue:42936",
+            "right": "issue:43240",
+            "accept": false,
+            "reason": "Mask2Former ignore_value handling and kwargs not being forwarded to a loss function are distinct implementation issues in different code paths."
+          },
+          {
+            "left": "issue:42936",
+            "right": "issue:44031",
+            "accept": false,
+            "reason": "These are unrelated: one is a model-specific ignore_value bug, the other is a tokenizer regex warning after a version change."
+          },
+          {
+            "left": "issue:42936",
+            "right": "issue:43813",
+            "accept": false,
+            "reason": "A Mask2Former loss/config bug and a typo in a quantization attribute are not the same underlying defect."
+          },
+          {
+            "left": "issue:42936",
+            "right": "issue:43668",
+            "accept": false,
+            "reason": "Different models and different problem classes: unused ignore_value versus an incorrect type hint."
+          },
+          {
+            "left": "issue:43240",
+            "right": "issue:43668",
+            "accept": false,
+            "reason": "Forwarding kwargs to a loss function and correcting a config type hint are unrelated fixes."
+          },
+          {
+            "left": "issue:43668",
+            "right": "issue:44031",
+            "accept": false,
+            "reason": "A ModernBERTConfig annotation issue does not match a tokenizer regex warning regression."
+          },
+          {
+            "left": "issue:43668",
+            "right": "issue:43813",
+            "accept": false,
+            "reason": "These only share the fact that both are small correctness issues; one is a type-hint typo, the other is a symbol-name typo in quantization code."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 7,
+      "component_count": 16,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-41084-2",
+      "nodes": [
+        "issue:41084",
+        "issue:43037"
+      ],
+      "soft_pairs": [
+        "issue:41084|issue:43037"
+      ],
+      "prepared_review_unit_hash": "d285e80d4f39570046f2f340b8ad8458e2a86dfea4ef0aad543c19edf64ffbb0",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 1479,
+        "estimated_input_tokens": 370,
+        "estimated_eval_tokens": 996
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 1479,
+        "estimated_input_tokens": 370,
+        "estimated_eval_tokens": 996
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "The two issues are not duplicates: one is a generic block-decoding request, while the other is a specific request for DeepSeek v3.2 support. There is no shared underlying bug or change here.",
+        "confidence": 0.97,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": null,
+        "best_issue_reason": null,
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:41084",
+            "right": "issue:43037",
+            "accept": false,
+            "reason": "Different topics and scopes: block decoding vs model/version support. The titles and available metadata do not indicate the same underlying issue."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 8,
+      "component_count": 16,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-43598-2",
+      "nodes": [
+        "issue:43598",
+        "issue:43600"
+      ],
+      "soft_pairs": [
+        "issue:43598|issue:43600"
+      ],
+      "prepared_review_unit_hash": "f52b10dfe50c835a1857ed7ee0257f8076630d6e3b10dfec219389e7e0e92e54",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 1584,
+        "estimated_input_tokens": 396,
+        "estimated_eval_tokens": 1048
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 1584,
+        "estimated_input_tokens": 396,
+        "estimated_eval_tokens": 1048
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "Both issues concern Trainer behavior with custom trainers, but they target different code paths: one is about `remove_unused_column` customizability, the other about FLOPS estimation. They are related in theme, not the same bug.",
+        "confidence": 0.81,
+        "canonical_issue_reason": "#43598 is the better canonical issue because it describes the broader Trainer customizability problem more directly and is more likely to serve as the umbrella for related follow-ups.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "#43598 is the best representative issue for the cluster since it frames the general Trainer customization concern more clearly; #43600 is a narrower instrumentation-specific concern.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:43598",
+            "right": "issue:43600",
+            "accept": false,
+            "reason": "Both mention custom trainers in Trainer, but `remove_unused_column` behavior and FLOPS estimation are different concrete problems and would not be merged as the same underlying issue."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 9,
+      "component_count": 16,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-43684-2",
+      "nodes": [
+        "issue:43684",
+        "issue:44405"
+      ],
+      "soft_pairs": [
+        "issue:43684|issue:44405"
+      ],
+      "prepared_review_unit_hash": "9a9403447ca689c684572748acf97cde2a3631de63d67c81f68089e8774c9db1",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 1612,
+        "estimated_input_tokens": 403,
+        "estimated_eval_tokens": 1062
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 1612,
+        "estimated_input_tokens": 403,
+        "estimated_eval_tokens": 1062
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "These are related auto-model feature requests, but they target different model families and different capabilities: Qwen3-Omni registration for AutoModel/AutoModelForConditionalGeneration versus Qwen3.5 AutoModelForSequenceClassification support. They do not look like duplicates.",
+        "confidence": 0.97,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": null,
+        "best_issue_reason": null,
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:43684",
+            "right": "issue:44405",
+            "accept": false,
+            "reason": "Different underlying changes: one is generic registration support for Qwen3-Omni, the other is task-specific sequence classification support for Qwen3.5/Qwen3_5Config. Similar area, but not the same bug or mergeable change."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 10,
+      "component_count": 16,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-43835-4",
+      "nodes": [
+        "issue:43835",
+        "issue:43845",
+        "issue:43846",
+        "issue:43859"
+      ],
+      "soft_pairs": [
+        "issue:43835|issue:43845",
+        "issue:43835|issue:43846",
+        "issue:43835|issue:43859",
+        "issue:43845|issue:43846",
+        "issue:43845|issue:43859",
+        "issue:43846|issue:43859"
+      ],
+      "prepared_review_unit_hash": "40795cb1c81b0677122066b8db925131904d7b8f7e10699348d9fd8f7ff7ef61",
+      "component_budget": {
+        "node_count": 4,
+        "item_count": 4,
+        "soft_pair_count": 6,
+        "serialized_chars": 4265,
+        "estimated_input_tokens": 1067,
+        "estimated_eval_tokens": 2390
+      },
+      "budget": {
+        "node_count": 4,
+        "item_count": 4,
+        "soft_pair_count": 6,
+        "serialized_chars": 4265,
+        "estimated_input_tokens": 1067,
+        "estimated_eval_tokens": 2390
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "This cluster is dominated by near-empty, generic issue titles with no actionable details. Three issues are effectively identical in form (\u201chuggingface \u201d), while one (\u201cFett\u201d) is unrelated. I would consolidate the huggingface reports under the earliest one, but the evidence is too thin to treat any pair as a true duplicate with confidence.",
+        "confidence": 0.63,
+        "canonical_issue_reason": "Issue 43845 is the earliest of the nearly identical \u2018huggingface \u2019 reports and is the most reasonable representative for the cluster.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "43845 is the best issue anchor because it is earliest among the repeated \u2018huggingface \u2019 submissions and matches the cluster\u2019s dominant pattern.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:43835",
+            "right": "issue:43845",
+            "accept": false,
+            "reason": "Only weak text similarity; \u2018Fett\u2019 and \u2018huggingface\u2019 do not show the same underlying bug or change."
+          },
+          {
+            "left": "issue:43835",
+            "right": "issue:43846",
+            "accept": false,
+            "reason": "Only weak text similarity; no concrete shared problem is described."
+          },
+          {
+            "left": "issue:43835",
+            "right": "issue:43859",
+            "accept": false,
+            "reason": "Only weak text similarity; unrelated-looking titles with no substantive issue details."
+          },
+          {
+            "left": "issue:43845",
+            "right": "issue:43846",
+            "accept": false,
+            "reason": "Very similar placeholders, but the bodies are too sparse to verify they are the same underlying issue."
+          },
+          {
+            "left": "issue:43845",
+            "right": "issue:43859",
+            "accept": false,
+            "reason": "Same broad title, but no concrete bug description or shared target to justify merging."
+          },
+          {
+            "left": "issue:43846",
+            "right": "issue:43859",
+            "accept": false,
+            "reason": "Same broad title, but insufficient evidence of one concrete shared defect."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 11,
+      "component_count": 16,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-44371-5",
+      "nodes": [
+        "issue:44371",
+        "issue:44418",
+        "issue:44572",
+        "issue:44573",
+        "issue:44937"
+      ],
+      "soft_pairs": [
+        "issue:44371|issue:44572",
+        "issue:44371|issue:44573",
+        "issue:44371|issue:44418",
+        "issue:44371|issue:44937",
+        "issue:44418|issue:44572",
+        "issue:44418|issue:44573",
+        "issue:44572|issue:44573",
+        "issue:44572|issue:44937",
+        "issue:44573|issue:44937"
+      ],
+      "prepared_review_unit_hash": "dc176460818cee6d63b5360de7889a2d1934e945d8a7aaae0d5d3966be4c376a",
+      "component_budget": {
+        "node_count": 5,
+        "item_count": 5,
+        "soft_pair_count": 9,
+        "serialized_chars": 5791,
+        "estimated_input_tokens": 1448,
+        "estimated_eval_tokens": 3152
+      },
+      "budget": {
+        "node_count": 5,
+        "item_count": 5,
+        "soft_pair_count": 9,
+        "serialized_chars": 5791,
+        "estimated_input_tokens": 1448,
+        "estimated_eval_tokens": 3152
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "All items appear to be unrelated low-information/spam issues; none has enough concrete evidence to be treated as a duplicate of another. No PRs are present.",
+        "confidence": 0.86,
+        "canonical_issue_reason": "Issue 44418 is the best representative only because its title is slightly more descriptive than the spam/placeholder titles; it still lacks enough substance to justify a true canonical duplicate target.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "Issue 44418 is the least spam-like and most human-readable issue title in the set, so it is the best single representative for triage purposes.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:44371",
+            "right": "issue:44572",
+            "accept": false,
+            "reason": "Both are placeholder spam titles with no body detail, targets, filenames, or other evidence of the same underlying issue."
+          },
+          {
+            "left": "issue:44371",
+            "right": "issue:44573",
+            "accept": false,
+            "reason": "Identical generic titles alone are not enough; there is no concrete evidence they describe the same bug or change."
+          },
+          {
+            "left": "issue:44371",
+            "right": "issue:44418",
+            "accept": false,
+            "reason": "A spam placeholder and a documentation-suggestion title do not indicate the same underlying report."
+          },
+          {
+            "left": "issue:44371",
+            "right": "issue:44937",
+            "accept": false,
+            "reason": "The titles are unrelated and there is no shared technical context to justify duplicate treatment."
+          },
+          {
+            "left": "issue:44418",
+            "right": "issue:44572",
+            "accept": false,
+            "reason": "A documentation enhancement request and a spam placeholder are not the same issue."
+          },
+          {
+            "left": "issue:44418",
+            "right": "issue:44573",
+            "accept": false,
+            "reason": "No concrete overlap beyond low-information metadata; this is not enough to accept a duplicate edge."
+          },
+          {
+            "left": "issue:44572",
+            "right": "issue:44573",
+            "accept": false,
+            "reason": "Matching spam-like titles do not establish the same underlying bug or change without additional context."
+          },
+          {
+            "left": "issue:44572",
+            "right": "issue:44937",
+            "accept": false,
+            "reason": "These titles are unrelated and there is no shared code-path or explicit linkage."
+          },
+          {
+            "left": "issue:44573",
+            "right": "issue:44937",
+            "accept": false,
+            "reason": "The records have no overlapping technical details; they should not be merged as duplicates."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 12,
+      "component_count": 16,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-45145-2",
+      "nodes": [
+        "issue:45145",
+        "issue:45151"
+      ],
+      "soft_pairs": [
+        "issue:45145|issue:45151"
+      ],
+      "prepared_review_unit_hash": "cf494af436be5616134f7b713cf8520a40d6e98ce87ce7f849d38300cf5fd9ce",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 1645,
+        "estimated_input_tokens": 412,
+        "estimated_eval_tokens": 1080
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 1645,
+        "estimated_input_tokens": 412,
+        "estimated_eval_tokens": 1080
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "The packet supports that issues 45145 and 45151 are highly similar, with the same title and a soft-similarity match. It does not establish that they are exact duplicates or that their bodies are identical. Issue 45145 is the slightly better canonical choice because it has an inbound reference, but the evidence is still limited.",
+        "confidence": 0.84,
+        "canonical_issue_reason": "45145 is the better canonical issue only by a small margin: it has the same title as 45151 and one inbound reference, which makes it a slightly stronger anchor for the cluster. The evidence does not prove exact duplication.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "45145 is the best global issue candidate because it is the most connected of the pair and has the same title as 45151. However, this is a soft cluster, so the choice is based on linkage strength rather than a confirmed duplicate relationship.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:45145",
+            "right": "issue:45151",
+            "accept": false,
+            "reason": "This is only a soft-similarity match with deterministic_accept=false. The titles match, but the packet does not provide enough evidence to treat them as the same underlying issue or an exact duplicate."
+          }
+        ]
+      },
+      "evaluator_result": {
+        "accept": false,
+        "feedback": "The summary overstates the evidence. The packet only supports that the two issues share the same title and have a soft-similarity match; it does not prove they are exact duplicates or have the same body text. The soft-edge accept is also too strong given the evidence: the pair is only a soft-similarity match and deterministic_accept is false. The inbound reference on 45145 can justify it as a slightly better canonical candidate, but not a strong or exact duplicate claim."
+      },
+      "evaluator_used": true,
+      "retried": true,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 13,
+      "component_count": 16,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-45295-2",
+      "nodes": [
+        "issue:45295",
+        "issue:45373"
+      ],
+      "soft_pairs": [],
+      "prepared_review_unit_hash": "f5fd95fada2f530ab0999e343c17c16266cde0c7182bb22509972ca44827cd49",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 0,
+        "serialized_chars": 1286,
+        "estimated_input_tokens": 322,
+        "estimated_eval_tokens": 900
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 0,
+        "serialized_chars": 1286,
+        "estimated_input_tokens": 322,
+        "estimated_eval_tokens": 900
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "These two issues describe the same missing Gemma 4 sequence-classification support; the later, more specific open issue is the better cluster representative.",
+        "confidence": 0.94,
+        "canonical_issue_reason": "issue #45373 is the clearest canonical tracker: it is open, explicitly names the missing class `Gemma4ForSequenceClassification`, and states the module-level gap relative to Gemma 2/3.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "issue #45373 best captures the concrete request and likely remaining work, while #45295 is a shorter earlier duplicate with less detail and is already closed.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": []
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 14,
+      "component_count": 16,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-45672-2",
+      "nodes": [
+        "issue:45672",
+        "issue:45674"
+      ],
+      "soft_pairs": [
+        "issue:45672|issue:45674"
+      ],
+      "prepared_review_unit_hash": "b043f18b803f39253da7363028f9c77ae154fb10f33764ceb312c669bdc73c1f",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 1668,
+        "estimated_input_tokens": 417,
+        "estimated_eval_tokens": 1090
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 1668,
+        "estimated_input_tokens": 417,
+        "estimated_eval_tokens": 1090
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "These are related only at a very broad level (4-bit inference / bitsandbytes), but they describe different bugs, different failure modes, and likely different fixes. They should not be merged as duplicates.",
+        "confidence": 0.94,
+        "canonical_issue_reason": "Issue 45674 is the more general and actionable report: it names the configuration trigger (`llm_int8_skip_modules`) and the resulting regression in 4-bit inference, making it a better representative issue for this cluster.",
+        "canonical_pr_reason": null,
+        "best_issue_reason": "Issue 45674 is the best overall representative because it describes a concrete, reusable configuration bug rather than a model-specific symptom, and the repro/failure mode is clearer.",
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:45672",
+            "right": "issue:45674",
+            "accept": false,
+            "reason": "Do not dedupe: 45672 is a Gemma4/audio-module `torch.finfo()` TypeError on uint8 weights, while 45674 is a BitsAndBytesConfig regression where `llm_int8_skip_modules` clears the default `lm_head` exclusion and triggers an AssertionError. Same broad area, different code paths and fixes."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 15,
+      "component_count": 16,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-45762-2",
+      "nodes": [
+        "issue:45762",
+        "issue:45763"
+      ],
+      "soft_pairs": [
+        "issue:45762|issue:45763"
+      ],
+      "prepared_review_unit_hash": "52b48b2ce0b96c9bf2b5c2bd05522b0a03c3757bc988decb84346ac41de10946",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 1476,
+        "estimated_input_tokens": 369,
+        "estimated_eval_tokens": 994
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 1476,
+        "estimated_input_tokens": 369,
+        "estimated_eval_tokens": 994
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "These two issues do not look like duplicates of the same underlying bug or request; they are just two separate closed issues with different personal-name titles and dates and no supporting body text.",
+        "confidence": 0.98,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": null,
+        "best_issue_reason": null,
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:45762",
+            "right": "issue:45763",
+            "accept": false,
+            "reason": "Different titles/names and dates, no body, no shared bug description or explicit linkage; this does not indicate the same underlying issue."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "issue",
+      "component_index": 16,
+      "component_count": 16,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-45764-2",
+      "nodes": [
+        "issue:45764",
+        "issue:45765"
+      ],
+      "soft_pairs": [
+        "issue:45764|issue:45765"
+      ],
+      "prepared_review_unit_hash": "3d886d2092974221bd5effb3ded5b764e5381609f7336d09af99a1fb6494aa52",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 1481,
+        "estimated_input_tokens": 371,
+        "estimated_eval_tokens": 998
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 1481,
+        "estimated_input_tokens": 371,
+        "estimated_eval_tokens": 998
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "The two closed issues appear unrelated: both are bare personal-name/date titles with no body or shared technical context, so they should not be clustered as duplicates.",
+        "confidence": 0.98,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": null,
+        "best_issue_reason": null,
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "issue:45764",
+            "right": "issue:45765",
+            "accept": false,
+            "reason": "The titles refer to different people and different dates, with no shared bug report, code path, or issue context."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 1,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "llm",
+      "cache_hit": false,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-39895-13",
+      "nodes": [
+        "pull_request:39895",
+        "pull_request:41855",
+        "pull_request:42668",
+        "pull_request:43251",
+        "pull_request:43254",
+        "pull_request:43424",
+        "pull_request:43823",
+        "pull_request:43995",
+        "pull_request:44827",
+        "pull_request:45170",
+        "pull_request:45213",
+        "pull_request:45655",
+        "pull_request:45795"
+      ],
+      "soft_pairs": [
+        "pull_request:39895|pull_request:43424",
+        "pull_request:41855|pull_request:43424",
+        "pull_request:41855|pull_request:43823",
+        "pull_request:42668|pull_request:43823",
+        "pull_request:43251|pull_request:43823",
+        "pull_request:43251|pull_request:44827",
+        "pull_request:43424|pull_request:44827",
+        "pull_request:43424|pull_request:43823",
+        "pull_request:43823|pull_request:45655",
+        "pull_request:43823|pull_request:43995",
+        "pull_request:43995|pull_request:44827",
+        "pull_request:43995|pull_request:45213",
+        "pull_request:44827|pull_request:45170",
+        "pull_request:44827|pull_request:45795",
+        "pull_request:44827|pull_request:45655",
+        "pull_request:44827|pull_request:45213",
+        "pull_request:43823|pull_request:45213",
+        "pull_request:43823|pull_request:45795",
+        "pull_request:43251|pull_request:43254"
+      ],
+      "prepared_review_unit_hash": "89c73c8ba9e08bb158f968e0590ca6071f3427f941b3b30a742a310d6c241f26",
+      "component_budget": {
+        "node_count": 13,
+        "item_count": 13,
+        "soft_pair_count": 19,
+        "serialized_chars": 35208,
+        "estimated_input_tokens": 8802,
+        "estimated_eval_tokens": 17860
+      },
+      "budget": {
+        "node_count": 13,
+        "item_count": 13,
+        "soft_pair_count": 19,
+        "serialized_chars": 35208,
+        "estimated_input_tokens": 8802,
+        "estimated_eval_tokens": 17860
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "Mostly unrelated PRs with one real duplicate-like pair in `fixed_cross_entropy`. The loss-helper pair (43251/43254) is the only strong merge candidate; the rest are different models, docs, tests, or subsystems.",
+        "confidence": 0.94,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "PR 43251 is a reasonable canonical representative for the loss-helper pair because it makes the same `fixed_cross_entropy` change in `src/transformers/loss/loss_utils.py` as PR 43254 and both target issue 43240. The packet does not clearly justify a stronger preference beyond that.",
+        "best_issue_reason": null,
+        "best_pr_reason": "PR 43251 is a suitable global representative only for the one duplicate-like fix in this cluster: it and PR 43254 patch the same helper with the same supported kwargs in the same file, so either could stand in for that pair.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:39895",
+            "right": "pull_request:43424",
+            "accept": false,
+            "reason": "VideoPrism integration vs an ExecuTorch dynamic-shape test; different changes and no shared concrete bug."
+          },
+          {
+            "left": "pull_request:41855",
+            "right": "pull_request:43424",
+            "accept": false,
+            "reason": "Mistral tokenizer helper additions vs ExecuTorch export testing; unrelated code paths."
+          },
+          {
+            "left": "pull_request:41855",
+            "right": "pull_request:43823",
+            "accept": false,
+            "reason": "Tokenizer method fixes vs adding MobileLLM; same broad area only, not the same fix."
+          },
+          {
+            "left": "pull_request:42668",
+            "right": "pull_request:43823",
+            "accept": false,
+            "reason": "Processor/auto-registration work for different models; not the same underlying change."
+          },
+          {
+            "left": "pull_request:43251",
+            "right": "pull_request:43823",
+            "accept": false,
+            "reason": "Loss helper kwargs fix vs new model implementation; no common code-path problem."
+          },
+          {
+            "left": "pull_request:43251",
+            "right": "pull_request:44827",
+            "accept": false,
+            "reason": "Cross-entropy helper fix vs Mistral4 model/test changes; different bugs."
+          },
+          {
+            "left": "pull_request:43424",
+            "right": "pull_request:44827",
+            "accept": false,
+            "reason": "ExecuTorch dynamic-shape test vs Mistral4 fixes; same general testing theme only."
+          },
+          {
+            "left": "pull_request:43424",
+            "right": "pull_request:43823",
+            "accept": false,
+            "reason": "ExecuTorch export test vs MobileLLM addition; unrelated."
+          },
+          {
+            "left": "pull_request:43823",
+            "right": "pull_request:45655",
+            "accept": false,
+            "reason": "New model addition vs `cls.config` resolution fix; different subsystems."
+          },
+          {
+            "left": "pull_request:43823",
+            "right": "pull_request:43995",
+            "accept": false,
+            "reason": "MobileLLM implementation vs Falcon refactor; not duplicates."
+          },
+          {
+            "left": "pull_request:43995",
+            "right": "pull_request:44827",
+            "accept": false,
+            "reason": "Falcon output-collection refactor vs Mistral4 test/model work; different models and goals."
+          },
+          {
+            "left": "pull_request:43995",
+            "right": "pull_request:45213",
+            "accept": false,
+            "reason": "Falcon refactor vs a model-creation skill PR; completely different artifact types."
+          },
+          {
+            "left": "pull_request:44827",
+            "right": "pull_request:45170",
+            "accept": false,
+            "reason": "Mistral4 fixes vs CLIP-like `layrnorm` renames; different underlying issues."
+          },
+          {
+            "left": "pull_request:44827",
+            "right": "pull_request:45795",
+            "accept": false,
+            "reason": "Mistral4 work vs audio/video processor docs; not the same change."
+          },
+          {
+            "left": "pull_request:44827",
+            "right": "pull_request:45655",
+            "accept": false,
+            "reason": "Model-specific fixes vs core config-resolution ordering; no shared code-path bug."
+          },
+          {
+            "left": "pull_request:44827",
+            "right": "pull_request:45213",
+            "accept": false,
+            "reason": "Model-specific fixes vs repo skill/dataset PR; unrelated."
+          },
+          {
+            "left": "pull_request:43823",
+            "right": "pull_request:45213",
+            "accept": false,
+            "reason": "New model implementation vs draft model-creation skill/docs; same domain only."
+          },
+          {
+            "left": "pull_request:43823",
+            "right": "pull_request:45795",
+            "accept": false,
+            "reason": "Model implementation vs docs for audio/video processors; related topic, not a duplicate."
+          },
+          {
+            "left": "pull_request:43251",
+            "right": "pull_request:43254",
+            "accept": true,
+            "reason": "Both PRs change `src/transformers/loss/loss_utils.py` in the same `fixed_cross_entropy` helper, add the same kwargs (`weight`, `label_smoothing`), and target the same issue 43240. They are the same underlying fix and could be merged into one PR."
+          }
+        ]
+      },
+      "evaluator_result": {
+        "accept": false,
+        "feedback": "Reject: the canonical PR rationale overstates the packet. PR 43254 also has explicit issue target 43240, so it does not \"lack the issue linkage.\" The claim that 43251 is broader/cleaner is also not directly grounded in the packet. The accepted duplicate pair itself looks mergeable, but the summary should be more conservative."
+      },
+      "evaluator_used": true,
+      "retried": true,
+      "accepted_nontrivial_soft_edge": true,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 2,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "llm",
+      "cache_hit": false,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-40546-2",
+      "nodes": [
+        "pull_request:40546",
+        "pull_request:42781"
+      ],
+      "soft_pairs": [
+        "pull_request:40546|pull_request:42781"
+      ],
+      "prepared_review_unit_hash": "582e4f4f7ae86a8f5aa999749fca767e875d77206a05f6c275d78d4defc91332",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 7226,
+        "estimated_input_tokens": 1807,
+        "estimated_eval_tokens": 3870
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 7226,
+        "estimated_input_tokens": 1807,
+        "estimated_eval_tokens": 3870
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "Both PRs are about VibeVoice, but they target different deliverables: one adds the base VibeVoice model integration, while the other adds VibeVoice Realtime and an acoustic tokenizer with additional generation/runtime changes. They overlap on shared model scaffolding, but not on the same concrete change.",
+        "confidence": 0.93,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "PR #40546 is the better canonical PR because it is the foundational VibeVoice implementation and broader base integration; the other PR builds a separate realtime variant on top of it.",
+        "best_issue_reason": null,
+        "best_pr_reason": "PR #40546 is the best representative of the cluster because it covers the core model addition and is the more central, non-draft implementation.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:40546",
+            "right": "pull_request:42781",
+            "accept": false,
+            "reason": "Reject: they share the VibeVoice family and some files, but one is the base model integration and the other is a realtime/acoustic-tokenizer extension with different code paths and scope. They do not look like the same mergeable fix/change."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 3,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "llm",
+      "cache_hit": false,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-40755-2",
+      "nodes": [
+        "pull_request:40755",
+        "pull_request:41224"
+      ],
+      "soft_pairs": [
+        "pull_request:40755|pull_request:41224"
+      ],
+      "prepared_review_unit_hash": "288f99592cd9551e0c534578064187a69297a2b7027bcd7d01c25f029b5f9df3",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 4676,
+        "estimated_input_tokens": 1169,
+        "estimated_eval_tokens": 2594
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 4676,
+        "estimated_input_tokens": 1169,
+        "estimated_eval_tokens": 2594
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "The two pull requests are unrelated: one adds TimesFM covariate forecasting, the other adds DINOv3 image classification support. They touch different models, code paths, and documentation, so this is not a duplicate cluster.",
+        "confidence": 0.99,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": null,
+        "best_issue_reason": null,
+        "best_pr_reason": "PR #40755 is the broader, more substantial feature addition in this cluster, but it is not a duplicate of #41224 and should not be treated as a canonical match.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:40755",
+            "right": "pull_request:41224",
+            "accept": false,
+            "reason": "Different models and features (TimesFM covariates vs. DINOv3 image classification); no shared underlying bug or mergeable change."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 4,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-41356-2",
+      "nodes": [
+        "pull_request:41356",
+        "pull_request:44339"
+      ],
+      "soft_pairs": [
+        "pull_request:41356|pull_request:44339"
+      ],
+      "prepared_review_unit_hash": "aead1308c167b0bae6ff11bf9358340baa5e76f816618383418b505a10a9cd01",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 6050,
+        "estimated_input_tokens": 1513,
+        "estimated_eval_tokens": 3282
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 6050,
+        "estimated_input_tokens": 1513,
+        "estimated_eval_tokens": 3282
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "Two DEIMv2 pull requests target the same model family, but they are not the same change: PR 41356 is an early/basic model + image processor add, while PR 44339 is a much broader follow-up that adds loss code, auto-mappings, conversion tooling, modular files, and more integration work.",
+        "confidence": 0.91,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "PR 44339 is the more complete and representative DEIMv2 implementation; PR 41356 looks like an earlier partial version of the same model bring-up.",
+        "best_issue_reason": null,
+        "best_pr_reason": "PR 44339 is the strongest single artifact for the cluster because it subsumes the basic model work and includes the broader integration needed for a full Transformers addition.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:41356",
+            "right": "pull_request:44339",
+            "accept": false,
+            "reason": "Same model area, but not the same concrete change: 44339 expands the implementation substantially and adds new code paths beyond the basic PR 41356. They look like related iterations, not mergeable duplicates."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 5,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "llm",
+      "cache_hit": false,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-41797-3",
+      "nodes": [
+        "pull_request:41797",
+        "pull_request:45186",
+        "pull_request:45630"
+      ],
+      "soft_pairs": [
+        "pull_request:41797|pull_request:45186",
+        "pull_request:45186|pull_request:45630"
+      ],
+      "prepared_review_unit_hash": "742e11c30f1d1eb9c7d3b835a1b501e3ea127e469d11f7e9abdad35eda54cf1f",
+      "component_budget": {
+        "node_count": 3,
+        "item_count": 3,
+        "soft_pair_count": 2,
+        "serialized_chars": 10429,
+        "estimated_input_tokens": 2608,
+        "estimated_eval_tokens": 5472
+      },
+      "budget": {
+        "node_count": 3,
+        "item_count": 3,
+        "soft_pair_count": 2,
+        "serialized_chars": 10429,
+        "estimated_input_tokens": 2608,
+        "estimated_eval_tokens": 5472
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "These are three separate model-addition PRs (DeepSeek OCR, Isaac, Kimi2-6). They share generic Transformers boilerplate and registry/docs updates, but not the same underlying model or code-path change.",
+        "confidence": 0.97,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "No canonical PR: the cluster is heterogeneous, with each PR introducing a different model integration rather than duplicating the same change.",
+        "best_issue_reason": null,
+        "best_pr_reason": "PR 45630 is the most developed integration by discussion/review activity and scope, so it is the best representative of the cluster, but it is still not a duplicate of the others.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:41797",
+            "right": "pull_request:45186",
+            "accept": false,
+            "reason": "Different model integrations (DeepSeek OCR vs Isaac). The overlap is limited to standard new-model scaffolding in docs, auto mappings, and utility updates, not the same bug or feature."
+          },
+          {
+            "left": "pull_request:45186",
+            "right": "pull_request:45630",
+            "accept": false,
+            "reason": "Different model integrations (Isaac vs Kimi2-6). Shared registry/docs churn is expected for new models and does not indicate a common underlying code-path change."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 6,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "llm",
+      "cache_hit": false,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-42975-2",
+      "nodes": [
+        "pull_request:42975",
+        "pull_request:42976"
+      ],
+      "soft_pairs": [
+        "pull_request:42975|pull_request:42976"
+      ],
+      "prepared_review_unit_hash": "a3fc123a430a9441552d609af6d627312dbe0c7aba616f0b8d026cc87d346c2a",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 5935,
+        "estimated_input_tokens": 1484,
+        "estimated_eval_tokens": 3224
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 5935,
+        "estimated_input_tokens": 1484,
+        "estimated_eval_tokens": 3224
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "Two workflow-maintenance PRs that both upgrade GitHub Actions, but they target different actions and different goals. They are related by theme, not the same underlying change, so they should not be deduplicated.",
+        "confidence": 0.95,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "No duplicate canonical PR: the two PRs address different action upgrades and are not the same mergeable change.",
+        "best_issue_reason": null,
+        "best_pr_reason": "PR 42975 is the better representative of the cluster because it is the broader workflow-actions compatibility update (checkout/cache/upload-artifact) and explicitly targets Node 24 compatibility.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:42975",
+            "right": "pull_request:42976",
+            "accept": false,
+            "reason": "Both modify GitHub Actions workflows, but 42975 upgrades checkout/cache/upload-artifact for Node 24 compatibility while 42976 bumps docker/build-push-action to v6. Different actions, different concrete change, not the same underlying bug/fix."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 7,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "llm",
+      "cache_hit": false,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-42978-7",
+      "nodes": [
+        "pull_request:42978",
+        "pull_request:43448",
+        "pull_request:43451",
+        "pull_request:43973",
+        "pull_request:44601",
+        "pull_request:45490",
+        "pull_request:45838"
+      ],
+      "soft_pairs": [
+        "pull_request:42978|pull_request:43448",
+        "pull_request:42978|pull_request:43451",
+        "pull_request:42978|pull_request:45490",
+        "pull_request:43448|pull_request:43451",
+        "pull_request:43448|pull_request:45490",
+        "pull_request:43451|pull_request:45490",
+        "pull_request:43973|pull_request:45838",
+        "pull_request:44601|pull_request:45838",
+        "pull_request:45490|pull_request:45838"
+      ],
+      "prepared_review_unit_hash": "ced02bef74514f2010a2792937efbb35b97d750f38f48d45a37a624086090049",
+      "component_budget": {
+        "node_count": 7,
+        "item_count": 7,
+        "soft_pair_count": 9,
+        "serialized_chars": 23514,
+        "estimated_input_tokens": 5879,
+        "estimated_eval_tokens": 12014
+      },
+      "budget": {
+        "node_count": 7,
+        "item_count": 7,
+        "soft_pair_count": 9,
+        "serialized_chars": 23514,
+        "estimated_input_tokens": 5879,
+        "estimated_eval_tokens": 12014
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "These PRs are mostly unrelated model additions or infrastructure work; the high soft-similarity scores come from shared Hugging Face boilerplate, docs, and auto-registration files rather than the same underlying change. I found no true duplicate PR pair in this cluster.",
+        "confidence": 0.97,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "No canonical PR for dedupe: the cluster spans separate changes for ViT NEPA, Molmo, Molmo2, LFM2 audio, distributed pipeline-parallel loading, CTSM, and PP-OCRv6, with only template/docs overlap.",
+        "best_issue_reason": null,
+        "best_pr_reason": "No PR is a strong canonical duplicate target here; if a representative must be chosen, 43451 is the most substantial standalone model-addition PR, but it is not a duplicate of the others.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:42978",
+            "right": "pull_request:43448",
+            "accept": false,
+            "reason": "Different model additions (ViT NEPA vs Molmo); overlap is only in docs and auto-mapping boilerplate."
+          },
+          {
+            "left": "pull_request:42978",
+            "right": "pull_request:43451",
+            "accept": false,
+            "reason": "Different model families (ViT NEPA vs Molmo2) and distinct implementation paths; shared files are generic registration/docs."
+          },
+          {
+            "left": "pull_request:42978",
+            "right": "pull_request:45490",
+            "accept": false,
+            "reason": "ViT NEPA model addition and CTSM time-series model addition are unrelated changes."
+          },
+          {
+            "left": "pull_request:43448",
+            "right": "pull_request:43451",
+            "accept": false,
+            "reason": "Molmo and Molmo2 are related by name but are separate model releases with different code paths, not the same bug/change."
+          },
+          {
+            "left": "pull_request:43448",
+            "right": "pull_request:45490",
+            "accept": false,
+            "reason": "Molmo vs CTSM are unrelated model additions; only broad docs/auto files overlap."
+          },
+          {
+            "left": "pull_request:43451",
+            "right": "pull_request:45490",
+            "accept": false,
+            "reason": "Molmo2 and CTSM touch different subsystems and model implementations, so they are not duplicates."
+          },
+          {
+            "left": "pull_request:43973",
+            "right": "pull_request:45838",
+            "accept": false,
+            "reason": "Lfm2 audio support and PP-OCRv6 text recognition support are unrelated model families with different code paths."
+          },
+          {
+            "left": "pull_request:44601",
+            "right": "pull_request:45838",
+            "accept": false,
+            "reason": "Distributed pipeline-parallel loading infrastructure is unrelated to PP-OCRv6 model support; shared similarity is incidental."
+          },
+          {
+            "left": "pull_request:45490",
+            "right": "pull_request:45838",
+            "accept": false,
+            "reason": "CTSM time-series model support and PP-OCRv6 OCR model support are unrelated changes."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 8,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "llm",
+      "cache_hit": false,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-43291-3",
+      "nodes": [
+        "pull_request:43291",
+        "pull_request:43665",
+        "pull_request:44815"
+      ],
+      "soft_pairs": [
+        "pull_request:43291|pull_request:43665",
+        "pull_request:43291|pull_request:44815",
+        "pull_request:43665|pull_request:44815"
+      ],
+      "prepared_review_unit_hash": "cf40429a6a401b024316c5456f4429663061c9d1a3210e5e7a54e1a166b2f7fd",
+      "component_budget": {
+        "node_count": 3,
+        "item_count": 3,
+        "soft_pair_count": 3,
+        "serialized_chars": 7093,
+        "estimated_input_tokens": 1774,
+        "estimated_eval_tokens": 3804
+      },
+      "budget": {
+        "node_count": 3,
+        "item_count": 3,
+        "soft_pair_count": 3,
+        "serialized_chars": 7093,
+        "estimated_input_tokens": 1774,
+        "estimated_eval_tokens": 3804
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "All three are pull requests, but they do not look like the same underlying change. PR 44815 is the only substantive bug-fix PR; 43291 is a Whisper tokenizer/test change, and 43665 is just placeholder test edits. No pair is close enough to merge as a duplicate.",
+        "confidence": 0.68,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "PR 44815 is the strongest canonical candidate because it contains the only real implementation fix, spanning model loading and FP8 dequantization logic. The other two are test-focused and do not match its code path.",
+        "best_issue_reason": null,
+        "best_pr_reason": "PR 44815 is the best representative of the cluster: it has the clearest concrete bug fix, the broadest real code impact, and the most substantive diff. The others look like ancillary or unrelated test churn.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:43291",
+            "right": "pull_request:43665",
+            "accept": false,
+            "reason": "Both touch tests, but 43291 also changes Whisper tokenizer behavior while 43665 is only placeholder assertions in unrelated model tests; they are not the same bug or change."
+          },
+          {
+            "left": "pull_request:43291",
+            "right": "pull_request:44815",
+            "accept": false,
+            "reason": "43291 is a Whisper decoding/test fix, whereas 44815 is a dequantization/model-loading fix in a different subsystem; no shared concrete code-path problem."
+          },
+          {
+            "left": "pull_request:43665",
+            "right": "pull_request:44815",
+            "accept": false,
+            "reason": "43665 is clearly synthetic test churn, not a real implementation fix, so it is not the same underlying change as 44815."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 9,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "llm",
+      "cache_hit": false,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-43488-2",
+      "nodes": [
+        "pull_request:43488",
+        "pull_request:43532"
+      ],
+      "soft_pairs": [
+        "pull_request:43488|pull_request:43532"
+      ],
+      "prepared_review_unit_hash": "8fe2f04d1e47126838a5f89e17d0e57c6ce649a512df17a8da4dcfb024707677",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 4563,
+        "estimated_input_tokens": 1141,
+        "estimated_eval_tokens": 2538
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 4563,
+        "estimated_input_tokens": 1141,
+        "estimated_eval_tokens": 2538
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "These are not duplicates: PR #43488 is a synthetic repo-bot/checking change around setup/versioning and a Vit test, while PR #43532 rewires CI/workflow logic and touches several files. They only share a test filename, not the same underlying fix.",
+        "confidence": 0.96,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "PR #43532 is the better representative of the cluster because it is the more substantial, real change; PR #43488 is a placeholder-style bot check and does not match the same code path or change intent.",
+        "best_issue_reason": null,
+        "best_pr_reason": "PR #43532 is the most suitable overall candidate: it has broader, coherent changes across CI and helper code, whereas #43488 is a small synthetic setup/test tweak.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:43488",
+            "right": "pull_request:43532",
+            "accept": false,
+            "reason": "Reject: the overlap is only a shared Vit test file. #43488 changes packaging/versioning and a test, while #43532 changes workflow triggers and related CI utilities; they do not fix the same concrete bug or belong in one PR."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 10,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "llm",
+      "cache_hit": false,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-43612-2",
+      "nodes": [
+        "pull_request:43612",
+        "pull_request:43613"
+      ],
+      "soft_pairs": [
+        "pull_request:43612|pull_request:43613"
+      ],
+      "prepared_review_unit_hash": "7de38e7685f087524a82468bcff6beeae8336e22406c4b289d9dbbc74dbfac70",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 6082,
+        "estimated_input_tokens": 1521,
+        "estimated_eval_tokens": 3298
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 6082,
+        "estimated_input_tokens": 1521,
+        "estimated_eval_tokens": 3298
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "The two PRs are structurally similar pipeline additions, but they implement different features: Promptable Concept Segmentation vs Promptable Visual Segmentation. They share documentation/metadata scaffolding, yet the model docs, pipeline modules, and tests target different code paths and use different underlying model families.",
+        "confidence": 0.96,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": null,
+        "best_issue_reason": null,
+        "best_pr_reason": null,
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:43612",
+            "right": "pull_request:43613",
+            "accept": false,
+            "reason": "Reject: these are not the same underlying bug or change. They add different pipelines with different task semantics and different model/test coverage, so they would not plausibly be merged into one PR."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 11,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "llm",
+      "cache_hit": false,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-43651-2",
+      "nodes": [
+        "pull_request:43651",
+        "pull_request:43663"
+      ],
+      "soft_pairs": [
+        "pull_request:43651|pull_request:43663"
+      ],
+      "prepared_review_unit_hash": "f062008cef8e56e673881cb4fb6961fc4641be5f95147c5ce3f00211e7b9ee6f",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 4292,
+        "estimated_input_tokens": 1073,
+        "estimated_eval_tokens": 2402
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 4292,
+        "estimated_input_tokens": 1073,
+        "estimated_eval_tokens": 2402
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "These two PRs both touch `trainer.py` and add extension hooks for custom trainers, but they solve different problems: gradient-accumulation loss scaling vs. column filtering based on the model signature. They should not be deduplicated.",
+        "confidence": 0.98,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "No canonical PR: the changes target different trainer behaviors and are not the same underlying fix.",
+        "best_issue_reason": null,
+        "best_pr_reason": "Neither PR is a good global representative for a duplicate cluster because they are unrelated feature changes, not two versions of the same fix.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:43651",
+            "right": "pull_request:43663",
+            "accept": false,
+            "reason": "Both modify `trainer.py`, but one adds `_loss_is_scaled_for_ga` for loss normalization and the other adds `_get_signature_columns` for dataset column filtering; they address different code paths and are not mergeable as a single duplicate PR."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 12,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "llm",
+      "cache_hit": false,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-43656-3",
+      "nodes": [
+        "pull_request:43656",
+        "pull_request:43836",
+        "pull_request:43842"
+      ],
+      "soft_pairs": [
+        "pull_request:43656|pull_request:43842",
+        "pull_request:43836|pull_request:43842"
+      ],
+      "prepared_review_unit_hash": "f53ec09360f9bbba87f17d58c9446b6ebc6bb5503fa1c4f143337f326455a17f",
+      "component_budget": {
+        "node_count": 3,
+        "item_count": 3,
+        "soft_pair_count": 2,
+        "serialized_chars": 5993,
+        "estimated_input_tokens": 1499,
+        "estimated_eval_tokens": 3254
+      },
+      "budget": {
+        "node_count": 3,
+        "item_count": 3,
+        "soft_pair_count": 2,
+        "serialized_chars": 5993,
+        "estimated_input_tokens": 1499,
+        "estimated_eval_tokens": 3254
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "43842 is the cleanest representative of the serve.py TypeAdapter NameError fix. 43836 is a plausible duplicate/near-duplicate because it targets the same file and the same issue area, but 43656 is broader and less directly grounded as the same fix.",
+        "confidence": 0.78,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "PR 43842 is the most targeted patch for the TypeAdapter NameError in `src/transformers/cli/serve.py`, with a minimal diff and a direct issue link.",
+        "best_issue_reason": null,
+        "best_pr_reason": "PR 43842 is the best cluster representative because it fixes the concrete TypeAdapter annotation problem with the least extra surface area.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:43656",
+            "right": "pull_request:43842",
+            "accept": false,
+            "reason": "Both touch `serve.py`, but 43656\u2019s `__future__`/annotation rewrite is broader and its explicit issue target differs, so the evidence is not strong enough to call it the same concrete fix as 43842."
+          },
+          {
+            "left": "pull_request:43836",
+            "right": "pull_request:43842",
+            "accept": true,
+            "reason": "These are directly aligned on the same `serve.py` TypeAdapter NameError path, and 43836\u2019s broader annotation cleanup still appears to be addressing the same runtime typing issue."
+          }
+        ]
+      },
+      "evaluator_result": {
+        "accept": false,
+        "feedback": "The summary overstates what the packet supports. All three PRs touch the same file, but the evidence does not establish that they are all the same TypeAdapter NameError fix: 43656\u2019s preview shows a broader `__future__`/annotation rewrite and it has a different explicit issue target, so calling it a close variant of 43842 is too strong. The 43836\u219443842 pair is better supported, but the analyst still presents the cluster as one uniform bugfix without enough grounding. Please narrow the claims to what is directly evidenced and be more conservative on the soft-edge pair for 43656."
+      },
+      "evaluator_used": true,
+      "retried": true,
+      "accepted_nontrivial_soft_edge": true,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 13,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-43779-2",
+      "nodes": [
+        "pull_request:43779",
+        "pull_request:43816"
+      ],
+      "soft_pairs": [
+        "pull_request:43779|pull_request:43816"
+      ],
+      "prepared_review_unit_hash": "e7c674adcc031aa8883109a5dfc678b4407fb20b10764302092e384d031ba83a",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 4362,
+        "estimated_input_tokens": 1091,
+        "estimated_eval_tokens": 2438
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 4362,
+        "estimated_input_tokens": 1091,
+        "estimated_eval_tokens": 2438
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "Both pull requests appear to implement the same SwanLab integration change: adding run id/resume support in `integration_utils.py` for issue #43698, with one variant also expanding the docs. They are strong duplicate PR candidates.",
+        "confidence": 0.94,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "PR #43816 is the cleaner representative: it carries the same SwanLab id/resume support plus the accompanying documentation updates, making it the more complete version of the shared change.",
+        "best_issue_reason": null,
+        "best_pr_reason": "PR #43816 is the best global representative because it appears to include the full feature change and docs for SwanLab run id/resume support, while #43779 is the narrower implementation-focused variant.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:43779",
+            "right": "pull_request:43816",
+            "accept": true,
+            "reason": "Same underlying SwanLab callback enhancement for id/resume support in the same file and same issue target; the doc differences do not change the core code-path change, so these are duplicate PRs."
+          }
+        ]
+      },
+      "evaluator_result": {
+        "accept": true,
+        "feedback": "The summary is grounded in the packet: both PRs target issue #43698, touch the same file, and the previews indicate the same SwanLab id/resume support with #43816 adding docs. The duplicate/mergeable verdict is conservative enough given the shared target and overlapping code change."
+      },
+      "evaluator_used": true,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": true,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 14,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 2,
+      "status": "reviewed",
+      "reason": null,
+      "source": "llm",
+      "cache_hit": false,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-43996-28",
+      "nodes": [
+        "pull_request:43996",
+        "pull_request:43997",
+        "pull_request:44007",
+        "pull_request:44010",
+        "pull_request:44013",
+        "pull_request:44017",
+        "pull_request:44018",
+        "pull_request:44019",
+        "pull_request:44044",
+        "pull_request:44056",
+        "pull_request:44059",
+        "pull_request:44066",
+        "pull_request:44068",
+        "pull_request:44071",
+        "pull_request:44072",
+        "pull_request:44073",
+        "pull_request:44074",
+        "pull_request:44076",
+        "pull_request:44085",
+        "pull_request:44086",
+        "pull_request:44098",
+        "pull_request:44101",
+        "pull_request:44116",
+        "pull_request:44129",
+        "pull_request:44154",
+        "pull_request:44161",
+        "pull_request:44722"
+      ],
+      "soft_pairs": [
+        "pull_request:43996|pull_request:44085",
+        "pull_request:43996|pull_request:44044",
+        "pull_request:44066|pull_request:44085",
+        "pull_request:44007|pull_request:44072",
+        "pull_request:44072|pull_request:44722",
+        "pull_request:44066|pull_request:44072",
+        "pull_request:44013|pull_request:44044",
+        "pull_request:44066|pull_request:44086",
+        "pull_request:44066|pull_request:44071",
+        "pull_request:44018|pull_request:44068",
+        "pull_request:44066|pull_request:44068",
+        "pull_request:44018|pull_request:44066",
+        "pull_request:44019|pull_request:44722",
+        "pull_request:44086|pull_request:44722",
+        "pull_request:44019|pull_request:44085",
+        "pull_request:44018|pull_request:44019",
+        "pull_request:44019|pull_request:44071",
+        "pull_request:44068|pull_request:44116",
+        "pull_request:44068|pull_request:44722",
+        "pull_request:44018|pull_request:44116",
+        "pull_request:44071|pull_request:44722",
+        "pull_request:44018|pull_request:44086",
+        "pull_request:44018|pull_request:44085",
+        "pull_request:44068|pull_request:44086",
+        "pull_request:44071|pull_request:44085",
+        "pull_request:44059|pull_request:44086",
+        "pull_request:44085|pull_request:44086",
+        "pull_request:44059|pull_request:44722",
+        "pull_request:44098|pull_request:44101",
+        "pull_request:44018|pull_request:44071",
+        "pull_request:44017|pull_request:44161",
+        "pull_request:44017|pull_request:44722",
+        "pull_request:44017|pull_request:44073",
+        "pull_request:44068|pull_request:44071",
+        "pull_request:44071|pull_request:44086",
+        "pull_request:43997|pull_request:44073",
+        "pull_request:44017|pull_request:44071",
+        "pull_request:44073|pull_request:44101",
+        "pull_request:44059|pull_request:44071",
+        "pull_request:44073|pull_request:44161",
+        "pull_request:44056|pull_request:44101",
+        "pull_request:44059|pull_request:44068",
+        "pull_request:44018|pull_request:44059",
+        "pull_request:44101|pull_request:44161",
+        "pull_request:43997|pull_request:44017",
+        "pull_request:43997|pull_request:44056",
+        "pull_request:44071|pull_request:44076",
+        "pull_request:44074|pull_request:44101",
+        "pull_request:44056|pull_request:44161",
+        "pull_request:43997|pull_request:44018",
+        "pull_request:43997|pull_request:44161",
+        "pull_request:44074|pull_request:44161",
+        "pull_request:44018|pull_request:44161",
+        "pull_request:43997|pull_request:44071",
+        "pull_request:44010|pull_request:44071",
+        "pull_request:44010|pull_request:44073",
+        "pull_request:44018|pull_request:44129",
+        "pull_request:43997|pull_request:44074",
+        "pull_request:43997|pull_request:44129",
+        "pull_request:44071|pull_request:44129",
+        "pull_request:44010|pull_request:44018",
+        "pull_request:43996|pull_request:44161",
+        "pull_request:43996|pull_request:44017",
+        "pull_request:44010|pull_request:44044",
+        "pull_request:44044|pull_request:44074",
+        "pull_request:44066|pull_request:44722",
+        "pull_request:44085|pull_request:44722",
+        "pull_request:44085|pull_request:44129",
+        "pull_request:43996|pull_request:44007",
+        "pull_request:43996|pull_request:44129",
+        "pull_request:44044|pull_request:44129",
+        "pull_request:44044|pull_request:44154"
+      ],
+      "prepared_review_unit_hash": "394c9d1e321da9e348237be25e57d6ad2fb10665530a1ce559509a9fb7313438",
+      "component_budget": {
+        "node_count": 28,
+        "item_count": 28,
+        "soft_pair_count": 78,
+        "serialized_chars": 84904,
+        "estimated_input_tokens": 21226,
+        "estimated_eval_tokens": 42708
+      },
+      "budget": {
+        "node_count": 27,
+        "item_count": 27,
+        "soft_pair_count": 72,
+        "serialized_chars": 75099,
+        "estimated_input_tokens": 18775,
+        "estimated_eval_tokens": 37806
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "This cluster is mainly a shared output-tracing refactor stream across many models. The common umbrella is issue #43979, but only a few pairs are true duplicates at the PR level: same model/file and same concrete output-capture change.",
+        "confidence": 0.89,
+        "canonical_issue_reason": "Shared tracker #43979 is the umbrella issue for the standardized output-tracing migration across these PRs.",
+        "canonical_pr_reason": "PR 44066 is the best representative merge anchor: it explicitly targets #43979, makes a concrete GPT-J/CodeGen output-tracing refactor, and has the clearest direct duplicate overlap with 44722 and 44085.",
+        "best_issue_reason": "Issue #43979 is the global umbrella for this refactor line and is the natural canonical issue for deduping the related PRs.",
+        "best_pr_reason": "PR 44066 is the strongest single PR representative because it is explicit, concrete, and closest to the duplicated GPT-J follow-ups.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:43996",
+            "right": "pull_request:44085",
+            "accept": false,
+            "reason": "Different models/files; both are output-tracing refactors, but not the same code path."
+          },
+          {
+            "left": "pull_request:43996",
+            "right": "pull_request:44044",
+            "accept": false,
+            "reason": "Different model families and different concrete implementations."
+          },
+          {
+            "left": "pull_request:44066",
+            "right": "pull_request:44085",
+            "accept": true,
+            "reason": "Same GPT-J output-tracing code path in the same file; 44066 is a broader version of the same fix."
+          },
+          {
+            "left": "pull_request:44007",
+            "right": "pull_request:44072",
+            "accept": false,
+            "reason": "Different backbones and different files; only the refactor theme is shared."
+          },
+          {
+            "left": "pull_request:44072",
+            "right": "pull_request:44722",
+            "accept": false,
+            "reason": "EfficientNet versus GPT-J/CodeGen; unrelated concrete changes."
+          },
+          {
+            "left": "pull_request:44066",
+            "right": "pull_request:44072",
+            "accept": false,
+            "reason": "Different models and different code paths."
+          },
+          {
+            "left": "pull_request:44013",
+            "right": "pull_request:44044",
+            "accept": false,
+            "reason": "MobileNetV2 and DeBERTa-v2 are unrelated implementations."
+          },
+          {
+            "left": "pull_request:44066",
+            "right": "pull_request:44086",
+            "accept": false,
+            "reason": "Different model code paths; same thematic refactor only."
+          },
+          {
+            "left": "pull_request:44066",
+            "right": "pull_request:44071",
+            "accept": false,
+            "reason": "GPT-J/CodeGen versus MPT; not the same bug or change."
+          },
+          {
+            "left": "pull_request:44018",
+            "right": "pull_request:44068",
+            "accept": true,
+            "reason": "Both are GPT-Neo output-tracing refactors on the same concrete code path."
+          },
+          {
+            "left": "pull_request:44066",
+            "right": "pull_request:44068",
+            "accept": false,
+            "reason": "GPT-J/CodeGen and GPT-Neo are separate implementations."
+          },
+          {
+            "left": "pull_request:44018",
+            "right": "pull_request:44066",
+            "accept": false,
+            "reason": "GPT-Neo versus GPT-J/CodeGen; same pattern, different code path."
+          },
+          {
+            "left": "pull_request:44019",
+            "right": "pull_request:44722",
+            "accept": false,
+            "reason": "ResNet and GPT-J/CodeGen are unrelated model implementations."
+          },
+          {
+            "left": "pull_request:44086",
+            "right": "pull_request:44722",
+            "accept": false,
+            "reason": "MGP-STR and GPT-J/CodeGen are different concrete changes."
+          },
+          {
+            "left": "pull_request:44019",
+            "right": "pull_request:44085",
+            "accept": false,
+            "reason": "ResNet versus GPT-J; only the refactor style matches."
+          },
+          {
+            "left": "pull_request:44018",
+            "right": "pull_request:44019",
+            "accept": false,
+            "reason": "GPT-Neo and ResNet are different models with different output paths."
+          },
+          {
+            "left": "pull_request:44019",
+            "right": "pull_request:44071",
+            "accept": false,
+            "reason": "Different model code paths."
+          },
+          {
+            "left": "pull_request:44068",
+            "right": "pull_request:44116",
+            "accept": false,
+            "reason": "GPT-Neo versus Flaubert; unrelated implementations."
+          },
+          {
+            "left": "pull_request:44068",
+            "right": "pull_request:44722",
+            "accept": false,
+            "reason": "GPT-Neo and GPT-J/CodeGen are different concrete refactors."
+          },
+          {
+            "left": "pull_request:44018",
+            "right": "pull_request:44116",
+            "accept": false,
+            "reason": "GPT-Neo and Flaubert are unrelated code paths."
+          },
+          {
+            "left": "pull_request:44071",
+            "right": "pull_request:44086",
+            "accept": false,
+            "reason": "MPT and MGP-STR are different model implementations."
+          },
+          {
+            "left": "pull_request:44071",
+            "right": "pull_request:44722",
+            "accept": false,
+            "reason": "MPT versus GPT-J/CodeGen; not mergeable as one PR."
+          },
+          {
+            "left": "pull_request:44059",
+            "right": "pull_request:44086",
+            "accept": false,
+            "reason": "GPT2 and MGP-STR are different concrete changes."
+          },
+          {
+            "left": "pull_request:44085",
+            "right": "pull_request:44086",
+            "accept": false,
+            "reason": "GPT-J and MGP-STR are unrelated code paths."
+          },
+          {
+            "left": "pull_request:44059",
+            "right": "pull_request:44722",
+            "accept": false,
+            "reason": "GPT2 versus GPT-J/CodeGen; same theme, different implementation."
+          },
+          {
+            "left": "pull_request:44098",
+            "right": "pull_request:44101",
+            "accept": false,
+            "reason": "ViLT and Flaubert/XLM are different model families."
+          },
+          {
+            "left": "pull_request:44018",
+            "right": "pull_request:44071",
+            "accept": false,
+            "reason": "GPT-Neo and MPT are unrelated concrete refactors."
+          },
+          {
+            "left": "pull_request:44017",
+            "right": "pull_request:44161",
+            "accept": false,
+            "reason": "SegFormer and LongT5 are unrelated model implementations."
+          },
+          {
+            "left": "pull_request:44017",
+            "right": "pull_request:44722",
+            "accept": false,
+            "reason": "SegFormer versus GPT-J/CodeGen; not the same bug or change."
+          },
+          {
+            "left": "pull_request:44017",
+            "right": "pull_request:44073",
+            "accept": false,
+            "reason": "SegFormer and VisualBERT are different code paths."
+          },
+          {
+            "left": "pull_request:44068",
+            "right": "pull_request:44071",
+            "accept": false,
+            "reason": "GPT-Neo and MPT are different models."
+          },
+          {
+            "left": "pull_request:44071",
+            "right": "pull_request:44086",
+            "accept": false,
+            "reason": "MPT and MGP-STR are unrelated implementations."
+          },
+          {
+            "left": "pull_request:44059",
+            "right": "pull_request:44071",
+            "accept": false,
+            "reason": "GPT2 and MPT are different code paths."
+          },
+          {
+            "left": "pull_request:44073",
+            "right": "pull_request:44101",
+            "accept": false,
+            "reason": "VisualBERT and Flaubert/XLM are different models."
+          },
+          {
+            "left": "pull_request:44056",
+            "right": "pull_request:44101",
+            "accept": false,
+            "reason": "MPNet and Flaubert/XLM are different implementations."
+          },
+          {
+            "left": "pull_request:44059",
+            "right": "pull_request:44068",
+            "accept": false,
+            "reason": "GPT2 and GPT-Neo are separate model code paths."
+          },
+          {
+            "left": "pull_request:44018",
+            "right": "pull_request:44059",
+            "accept": false,
+            "reason": "GPT-Neo versus GPT2; same refactor theme only."
+          },
+          {
+            "left": "pull_request:44101",
+            "right": "pull_request:44161",
+            "accept": false,
+            "reason": "Flaubert/XLM and LongT5 are unrelated changes."
+          },
+          {
+            "left": "pull_request:43997",
+            "right": "pull_request:44017",
+            "accept": false,
+            "reason": "RegNet and SegFormer are different model families."
+          },
+          {
+            "left": "pull_request:43997",
+            "right": "pull_request:44056",
+            "accept": false,
+            "reason": "RegNet and MPNet are unrelated."
+          },
+          {
+            "left": "pull_request:44071",
+            "right": "pull_request:44076",
+            "accept": false,
+            "reason": "MPT and ImageGPT are different concrete implementations."
+          },
+          {
+            "left": "pull_request:44074",
+            "right": "pull_request:44101",
+            "accept": false,
+            "reason": "TextNet and Flaubert/XLM are unrelated model code paths."
+          },
+          {
+            "left": "pull_request:44056",
+            "right": "pull_request:44161",
+            "accept": false,
+            "reason": "MPNet and LongT5 are different implementations."
+          },
+          {
+            "left": "pull_request:43997",
+            "right": "pull_request:44018",
+            "accept": false,
+            "reason": "RegNet and GPT-Neo are unrelated model code paths."
+          },
+          {
+            "left": "pull_request:43997",
+            "right": "pull_request:44161",
+            "accept": false,
+            "reason": "RegNet and LongT5 are different models."
+          },
+          {
+            "left": "pull_request:44074",
+            "right": "pull_request:44161",
+            "accept": false,
+            "reason": "TextNet and LongT5 are unrelated implementations."
+          },
+          {
+            "left": "pull_request:44018",
+            "right": "pull_request:44161",
+            "accept": false,
+            "reason": "GPT-Neo and LongT5 are different model code paths."
+          },
+          {
+            "left": "pull_request:43997",
+            "right": "pull_request:44071",
+            "accept": false,
+            "reason": "RegNet and MPT are unrelated."
+          },
+          {
+            "left": "pull_request:44010",
+            "right": "pull_request:44071",
+            "accept": false,
+            "reason": "SqueezeBERT and MPT are different models."
+          },
+          {
+            "left": "pull_request:44010",
+            "right": "pull_request:44073",
+            "accept": false,
+            "reason": "SqueezeBERT and VisualBERT are different concrete changes."
+          },
+          {
+            "left": "pull_request:44018",
+            "right": "pull_request:44129",
+            "accept": false,
+            "reason": "GPT-Neo and SpeechT5 are unrelated model implementations."
+          },
+          {
+            "left": "pull_request:44074",
+            "right": "pull_request:44161",
+            "accept": false,
+            "reason": "TextNet and LongT5 are different code paths."
+          },
+          {
+            "left": "pull_request:43997",
+            "right": "pull_request:44074",
+            "accept": false,
+            "reason": "RegNet and TextNet are unrelated models."
+          },
+          {
+            "left": "pull_request:43997",
+            "right": "pull_request:44129",
+            "accept": false,
+            "reason": "RegNet and SpeechT5 are different implementations."
+          },
+          {
+            "left": "pull_request:44071",
+            "right": "pull_request:44129",
+            "accept": false,
+            "reason": "MPT and SpeechT5 are unrelated."
+          },
+          {
+            "left": "pull_request:44010",
+            "right": "pull_request:44018",
+            "accept": false,
+            "reason": "SqueezeBERT and GPT-Neo are different model code paths."
+          },
+          {
+            "left": "pull_request:43996",
+            "right": "pull_request:44161",
+            "accept": false,
+            "reason": "FNet/CVT and LongT5 are unrelated."
+          },
+          {
+            "left": "pull_request:43996",
+            "right": "pull_request:44017",
+            "accept": false,
+            "reason": "FNet/CVT and SegFormer are different model families."
+          },
+          {
+            "left": "pull_request:44010",
+            "right": "pull_request:44044",
+            "accept": false,
+            "reason": "SqueezeBERT and DeBERTa-v2 are unrelated."
+          },
+          {
+            "left": "pull_request:44044",
+            "right": "pull_request:44074",
+            "accept": false,
+            "reason": "DeBERTa-v2 and TextNet are different concrete changes."
+          },
+          {
+            "left": "pull_request:44066",
+            "right": "pull_request:44722",
+            "accept": true,
+            "reason": "Same GPT-J/CodeGen output-tracing refactor; 44722 is a near-duplicate follow-up."
+          },
+          {
+            "left": "pull_request:44085",
+            "right": "pull_request:44722",
+            "accept": true,
+            "reason": "Both target the same GPT-J output-tracing code path and could be merged into one PR."
+          },
+          {
+            "left": "pull_request:44085",
+            "right": "pull_request:44129",
+            "accept": false,
+            "reason": "GPT-J and SpeechT5 are different implementations."
+          },
+          {
+            "left": "pull_request:43996",
+            "right": "pull_request:44007",
+            "accept": false,
+            "reason": "FNet/CVT and RegNet/ResNet/RT-DETR are unrelated."
+          },
+          {
+            "left": "pull_request:43996",
+            "right": "pull_request:44129",
+            "accept": false,
+            "reason": "FNet/CVT and SpeechT5 are different model code paths."
+          },
+          {
+            "left": "pull_request:44044",
+            "right": "pull_request:44129",
+            "accept": false,
+            "reason": "DeBERTa-v2 and SpeechT5 are unrelated implementations."
+          },
+          {
+            "left": "pull_request:44044",
+            "right": "pull_request:44154",
+            "accept": false,
+            "reason": "DeBERTa-v2 and ViTS are different concrete changes."
+          }
+        ]
+      },
+      "evaluator_result": {
+        "accept": true,
+        "feedback": "The cluster summary is grounded in the packet: #43979 is the shared issue target across the refactor stream, and the representative PR choice is supported by the GPT-J/CodeGen overlap. The soft-edge verdicts are conservative overall, and the true-accept pairs are limited to same-model/code-path cases that do look mergeable into a single bugfix PR."
+      },
+      "evaluator_used": true,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": true,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 14,
+      "component_count": 26,
+      "review_unit_index": 2,
+      "review_unit_count": 2,
+      "status": "reviewed",
+      "reason": null,
+      "source": "llm",
+      "cache_hit": false,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-43996-28",
+      "nodes": [
+        "pull_request:43997",
+        "pull_request:44010",
+        "pull_request:44024",
+        "pull_request:44056",
+        "pull_request:44073",
+        "pull_request:44074",
+        "pull_request:44076",
+        "pull_request:44129"
+      ],
+      "soft_pairs": [
+        "pull_request:44024|pull_request:44076",
+        "pull_request:44073|pull_request:44074",
+        "pull_request:44056|pull_request:44073",
+        "pull_request:44056|pull_request:44074",
+        "pull_request:43997|pull_request:44010",
+        "pull_request:44010|pull_request:44129"
+      ],
+      "prepared_review_unit_hash": "445c8ef2736f689f9e0062f64ab50a2fc6932c0e9fe647dae3963e1c0e91db7f",
+      "component_budget": {
+        "node_count": 28,
+        "item_count": 28,
+        "soft_pair_count": 78,
+        "serialized_chars": 84904,
+        "estimated_input_tokens": 21226,
+        "estimated_eval_tokens": 42708
+      },
+      "budget": {
+        "node_count": 8,
+        "item_count": 8,
+        "soft_pair_count": 6,
+        "serialized_chars": 17126,
+        "estimated_input_tokens": 4282,
+        "estimated_eval_tokens": 8820
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": true,
+      "analyst_result": {
+        "summary": "These PRs are all separate per-model migrations to standardized output-capturing/tracing patterns. They share a common refactor theme, but they do not appear to be the same concrete fix or mergeable into one PR.",
+        "confidence": 0.94,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "pull_request:44129 is the strongest representative: it has an explicit issue target, a concrete model-specific refactor, and the most clearly scoped change among the set.",
+        "best_issue_reason": null,
+        "best_pr_reason": "pull_request:44129 is the best overall PR because it is well-scoped, explicitly linked to an issue, and clearly documents a concrete output-tracing migration for one code path.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:44024",
+            "right": "pull_request:44076",
+            "accept": false,
+            "reason": "Both are standardized output-tracing refactors, but for different models (FocalNet vs ImageGPT) and different code paths; not the same bug/change."
+          },
+          {
+            "left": "pull_request:44073",
+            "right": "pull_request:44074",
+            "accept": false,
+            "reason": "VisualBert and TextNet are unrelated model implementations; these are parallel refactors, not one shared fix."
+          },
+          {
+            "left": "pull_request:44056",
+            "right": "pull_request:44073",
+            "accept": false,
+            "reason": "MPNet and VisualBert have different forward paths and output-recording setup; same refactor theme only, not duplicate work."
+          },
+          {
+            "left": "pull_request:44056",
+            "right": "pull_request:44074",
+            "accept": false,
+            "reason": "MPNet and TextNet are separate models with distinct output-capture changes; not the same underlying code-path problem."
+          },
+          {
+            "left": "pull_request:43997",
+            "right": "pull_request:44010",
+            "accept": false,
+            "reason": "RegNet and SqueezeBert are unrelated model modules; these PRs do not fix the same concrete issue."
+          },
+          {
+            "left": "pull_request:44010",
+            "right": "pull_request:44129",
+            "accept": false,
+            "reason": "SqueezeBert and SpeechT5 both touch output-tracing, but they affect different model internals and are not plausibly one merged PR."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 15,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "llm",
+      "cache_hit": false,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-43998-13",
+      "nodes": [
+        "pull_request:43998",
+        "pull_request:43999",
+        "pull_request:44000",
+        "pull_request:44001",
+        "pull_request:44002",
+        "pull_request:44003",
+        "pull_request:44004",
+        "pull_request:44025",
+        "pull_request:44026",
+        "pull_request:44027",
+        "pull_request:44028",
+        "pull_request:44029",
+        "pull_request:44030"
+      ],
+      "soft_pairs": [
+        "pull_request:43998|pull_request:44002",
+        "pull_request:43998|pull_request:44000",
+        "pull_request:43998|pull_request:44001",
+        "pull_request:43998|pull_request:44003",
+        "pull_request:43998|pull_request:44029",
+        "pull_request:43999|pull_request:44001",
+        "pull_request:43999|pull_request:44002",
+        "pull_request:43999|pull_request:44025",
+        "pull_request:43999|pull_request:44026",
+        "pull_request:43999|pull_request:44027",
+        "pull_request:44000|pull_request:44001",
+        "pull_request:44000|pull_request:44002",
+        "pull_request:44000|pull_request:44003",
+        "pull_request:44000|pull_request:44025",
+        "pull_request:44001|pull_request:44002",
+        "pull_request:44001|pull_request:44003",
+        "pull_request:44001|pull_request:44025",
+        "pull_request:44002|pull_request:44029",
+        "pull_request:44002|pull_request:44003",
+        "pull_request:44003|pull_request:44028",
+        "pull_request:44003|pull_request:44004",
+        "pull_request:43998|pull_request:44004",
+        "pull_request:44001|pull_request:44004",
+        "pull_request:44002|pull_request:44004",
+        "pull_request:44000|pull_request:44004",
+        "pull_request:44025|pull_request:44026",
+        "pull_request:44025|pull_request:44027",
+        "pull_request:44025|pull_request:44028",
+        "pull_request:44025|pull_request:44029",
+        "pull_request:44025|pull_request:44030",
+        "pull_request:44026|pull_request:44029",
+        "pull_request:44026|pull_request:44027",
+        "pull_request:44026|pull_request:44028",
+        "pull_request:44026|pull_request:44030",
+        "pull_request:44027|pull_request:44028",
+        "pull_request:44027|pull_request:44029",
+        "pull_request:44027|pull_request:44030",
+        "pull_request:44028|pull_request:44029",
+        "pull_request:44028|pull_request:44030",
+        "pull_request:44029|pull_request:44030"
+      ],
+      "prepared_review_unit_hash": "28c4cef5238fb6480721688abb6d4913ff90fd6e0deaaabfea8559a681ab8e12",
+      "component_budget": {
+        "node_count": 13,
+        "item_count": 13,
+        "soft_pair_count": 40,
+        "serialized_chars": 40179,
+        "estimated_input_tokens": 10045,
+        "estimated_eval_tokens": 20346
+      },
+      "budget": {
+        "node_count": 13,
+        "item_count": 13,
+        "soft_pair_count": 40,
+        "serialized_chars": 40179,
+        "estimated_input_tokens": 10045,
+        "estimated_eval_tokens": 20346
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "These PRs share a naming pattern and a broad \u201coutput tracing\u201d theme, but they are model-specific refactors in different code paths. I do not see true duplicate PRs here.",
+        "confidence": 0.95,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "No single true duplicate PR stands out; 44025 is the best representative anchor because it\u2019s a clear output-tracing refactor across two related model files, but it is still not the same change as the others.",
+        "best_issue_reason": null,
+        "best_pr_reason": "44025 is the strongest overall representative of the cluster\u2019s theme, with a broader two-file refactor footprint and central similarity to several others, even though it should not be treated as a duplicate of them.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:43998",
+            "right": "pull_request:44002",
+            "accept": false,
+            "reason": "Different models and code paths; same refactor theme only, not the same bug/change."
+          },
+          {
+            "left": "pull_request:43998",
+            "right": "pull_request:44000",
+            "accept": false,
+            "reason": "Separate model implementations; similarity is stylistic, not a shared concrete fix."
+          },
+          {
+            "left": "pull_request:43998",
+            "right": "pull_request:44001",
+            "accept": false,
+            "reason": "Different architecture and file; not the same underlying change."
+          },
+          {
+            "left": "pull_request:43998",
+            "right": "pull_request:44003",
+            "accept": false,
+            "reason": "Unrelated model code paths; only the refactor pattern matches."
+          },
+          {
+            "left": "pull_request:43998",
+            "right": "pull_request:44029",
+            "accept": false,
+            "reason": "Different model family; no evidence of one merged fix."
+          },
+          {
+            "left": "pull_request:43999",
+            "right": "pull_request:44001",
+            "accept": false,
+            "reason": "Different models; shared output-tracing cleanup is too broad to count as duplicate."
+          },
+          {
+            "left": "pull_request:43999",
+            "right": "pull_request:44002",
+            "accept": false,
+            "reason": "Separate model code paths with no concrete overlap beyond refactor style."
+          },
+          {
+            "left": "pull_request:43999",
+            "right": "pull_request:44025",
+            "accept": false,
+            "reason": "Different model families; not the same code-path problem."
+          },
+          {
+            "left": "pull_request:43999",
+            "right": "pull_request:44026",
+            "accept": false,
+            "reason": "Distinct encoder-decoder vs backbone model changes; not duplicate work."
+          },
+          {
+            "left": "pull_request:43999",
+            "right": "pull_request:44027",
+            "accept": false,
+            "reason": "Same generic tracing refactor, but different subsystem and behavior."
+          },
+          {
+            "left": "pull_request:44000",
+            "right": "pull_request:44001",
+            "accept": false,
+            "reason": "Different models; too generic to be a duplicate."
+          },
+          {
+            "left": "pull_request:44000",
+            "right": "pull_request:44002",
+            "accept": false,
+            "reason": "Separate code paths and model classes; only the theme overlaps."
+          },
+          {
+            "left": "pull_request:44000",
+            "right": "pull_request:44003",
+            "accept": false,
+            "reason": "No shared concrete bug or implementation target."
+          },
+          {
+            "left": "pull_request:44000",
+            "right": "pull_request:44025",
+            "accept": false,
+            "reason": "Different model families; same refactor motif is not enough."
+          },
+          {
+            "left": "pull_request:44001",
+            "right": "pull_request:44002",
+            "accept": false,
+            "reason": "Different architectures; not plausibly one PR."
+          },
+          {
+            "left": "pull_request:44001",
+            "right": "pull_request:44003",
+            "accept": false,
+            "reason": "No shared failing code-path, just similar cleanup."
+          },
+          {
+            "left": "pull_request:44001",
+            "right": "pull_request:44025",
+            "accept": false,
+            "reason": "Distinct model-specific output handling; not the same change."
+          },
+          {
+            "left": "pull_request:44002",
+            "right": "pull_request:44029",
+            "accept": false,
+            "reason": "Different model implementations and output semantics; not duplicate."
+          },
+          {
+            "left": "pull_request:44002",
+            "right": "pull_request:44003",
+            "accept": false,
+            "reason": "Same refactor category, but different concrete fixes."
+          },
+          {
+            "left": "pull_request:44003",
+            "right": "pull_request:44028",
+            "accept": false,
+            "reason": "Mamba vs SuperPoint are unrelated code paths; no duplicate evidence."
+          },
+          {
+            "left": "pull_request:44003",
+            "right": "pull_request:44004",
+            "accept": false,
+            "reason": "Different model internals and output shapes; not the same fix."
+          },
+          {
+            "left": "pull_request:43998",
+            "right": "pull_request:44004",
+            "accept": false,
+            "reason": "Backbone output tracing vs CodeGen cache/output tracing are different changes."
+          },
+          {
+            "left": "pull_request:44001",
+            "right": "pull_request:44004",
+            "accept": false,
+            "reason": "UnivNet and CodeGen are unrelated model code paths."
+          },
+          {
+            "left": "pull_request:44002",
+            "right": "pull_request:44004",
+            "accept": false,
+            "reason": "UperNet segmentation and CodeGen generation logic are not the same bug."
+          },
+          {
+            "left": "pull_request:44000",
+            "right": "pull_request:44004",
+            "accept": false,
+            "reason": "Different models and output contracts; only broad refactor similarity."
+          },
+          {
+            "left": "pull_request:44025",
+            "right": "pull_request:44026",
+            "accept": false,
+            "reason": "Depth Anything vs Vision Encoder-Decoder are separate model changes."
+          },
+          {
+            "left": "pull_request:44025",
+            "right": "pull_request:44027",
+            "accept": false,
+            "reason": "Different encoder-decoder stacks; not one concrete fix."
+          },
+          {
+            "left": "pull_request:44025",
+            "right": "pull_request:44028",
+            "accept": false,
+            "reason": "Depth estimation vs keypoint model; unrelated code paths."
+          },
+          {
+            "left": "pull_request:44025",
+            "right": "pull_request:44029",
+            "accept": false,
+            "reason": "Different model families; same tracing refactor pattern only."
+          },
+          {
+            "left": "pull_request:44025",
+            "right": "pull_request:44030",
+            "accept": false,
+            "reason": "Depth Anything and DPR have distinct implementations and behavior."
+          },
+          {
+            "left": "pull_request:44026",
+            "right": "pull_request:44029",
+            "accept": false,
+            "reason": "Vision encoder-decoder and RWKV are unrelated model code paths."
+          },
+          {
+            "left": "pull_request:44026",
+            "right": "pull_request:44027",
+            "accept": false,
+            "reason": "Different encoder-decoder models; too broad to merge as duplicates."
+          },
+          {
+            "left": "pull_request:44026",
+            "right": "pull_request:44028",
+            "accept": false,
+            "reason": "Distinct model architectures and outputs; not duplicate work."
+          },
+          {
+            "left": "pull_request:44026",
+            "right": "pull_request:44030",
+            "accept": false,
+            "reason": "Different model classes and code paths; only the refactor theme overlaps."
+          },
+          {
+            "left": "pull_request:44027",
+            "right": "pull_request:44028",
+            "accept": false,
+            "reason": "Speech encoder-decoder vs SuperPoint are not the same underlying change."
+          },
+          {
+            "left": "pull_request:44027",
+            "right": "pull_request:44029",
+            "accept": false,
+            "reason": "Unrelated model internals; shared tracing refactor is insufficient."
+          },
+          {
+            "left": "pull_request:44027",
+            "right": "pull_request:44030",
+            "accept": false,
+            "reason": "Different model families and behavior; not a duplicate."
+          },
+          {
+            "left": "pull_request:44028",
+            "right": "pull_request:44029",
+            "accept": false,
+            "reason": "SuperPoint and RWKV are separate subsystems with no shared concrete fix."
+          },
+          {
+            "left": "pull_request:44028",
+            "right": "pull_request:44030",
+            "accept": false,
+            "reason": "Different model implementations; no same code-path issue."
+          },
+          {
+            "left": "pull_request:44029",
+            "right": "pull_request:44030",
+            "accept": false,
+            "reason": "RWKV and DPR touch different architectures and output handling."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 16,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "llm",
+      "cache_hit": false,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-44662-3",
+      "nodes": [
+        "pull_request:44662",
+        "pull_request:45075",
+        "pull_request:45401"
+      ],
+      "soft_pairs": [
+        "pull_request:44662|pull_request:45401",
+        "pull_request:44662|pull_request:45075",
+        "pull_request:45075|pull_request:45401"
+      ],
+      "prepared_review_unit_hash": "1582ed2fc980c4e74af4b0db6df84c61d389c68623c8905a582892cfb7c903d4",
+      "component_budget": {
+        "node_count": 3,
+        "item_count": 3,
+        "soft_pair_count": 3,
+        "serialized_chars": 10912,
+        "estimated_input_tokens": 2728,
+        "estimated_eval_tokens": 5712
+      },
+      "budget": {
+        "node_count": 3,
+        "item_count": 3,
+        "soft_pair_count": 3,
+        "serialized_chars": 10912,
+        "estimated_input_tokens": 2728,
+        "estimated_eval_tokens": 5712
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "These are three separate model-addition PRs (PenguinVL, DeepSeek-OCR-2, and Voxtral TTS) that share the same Transformers boilerplate but do not fix or implement the same underlying change.",
+        "confidence": 0.96,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "No canonical PR: each PR introduces a different model integration with distinct configs, processors, model code, and tests.",
+        "best_issue_reason": null,
+        "best_pr_reason": "None stands out as a duplicate representative; the closest match is not a real duplicate because the PRs target different models and code paths.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:44662",
+            "right": "pull_request:45401",
+            "accept": false,
+            "reason": "Different models and different implementations (PenguinVL vs Voxtral TTS); shared auto/docs files are just standard model-registration boilerplate, not the same fix."
+          },
+          {
+            "left": "pull_request:44662",
+            "right": "pull_request:45075",
+            "accept": false,
+            "reason": "Both add new model support, but for different models (PenguinVL vs DeepSeek-OCR-2) with separate architecture, conversion, and test code."
+          },
+          {
+            "left": "pull_request:45075",
+            "right": "pull_request:45401",
+            "accept": false,
+            "reason": "Distinct model additions with different target problems and unique source trees; not mergeable as one PR for the same underlying change."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 17,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "llm",
+      "cache_hit": false,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-44979-2",
+      "nodes": [
+        "pull_request:44979",
+        "pull_request:45363"
+      ],
+      "soft_pairs": [
+        "pull_request:44979|pull_request:45363"
+      ],
+      "prepared_review_unit_hash": "db0324771111ae121e3998f1e5cbb1f1eac6293592181e1ad89e5f4517728939",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 4406,
+        "estimated_input_tokens": 1102,
+        "estimated_eval_tokens": 2460
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 4406,
+        "estimated_input_tokens": 1102,
+        "estimated_eval_tokens": 2460
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "Both PRs are about fusion-related APIs in Transformers, but they target different layers: #44979 adds a generic module-fusion API, while #45363 implements n-to-1 kernel fusion via `KernelConfig` and related integration changes. They are related, but not the same underlying change.",
+        "confidence": 0.89,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "#45363 is the more complete, user-facing fusion feature: it wires `KernelConfig` through model loading/integrations and includes an end-to-end example. #44979 is a separate foundational module-fusion API addition, not the same concrete change.",
+        "best_issue_reason": null,
+        "best_pr_reason": "#45363 is the better representative of the cluster because it contains the broader concrete implementation path for kernel fusion, touching core loading/integration code rather than only introducing a standalone API surface.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:44979",
+            "right": "pull_request:45363",
+            "accept": false,
+            "reason": "Related fusion work, but not the same bug or change. #44979 adds a generic module-fusion API in a new file, while #45363 implements kernel fusion through `KernelConfig` and integration changes. They would not plausibly be merged as one PR without collapsing distinct features."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 18,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "llm",
+      "cache_hit": false,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-45086-2",
+      "nodes": [
+        "pull_request:45086",
+        "pull_request:45317"
+      ],
+      "soft_pairs": [
+        "pull_request:45086|pull_request:45317"
+      ],
+      "prepared_review_unit_hash": "f75659896d3f7df0091b894c3bd8554367c49f17aa6da9486b53e9524e366814",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 4414,
+        "estimated_input_tokens": 1104,
+        "estimated_eval_tokens": 2464
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 4414,
+        "estimated_input_tokens": 1104,
+        "estimated_eval_tokens": 2464
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "Both pull requests fix the same AttributeError in `_patch_mistral_regex` by switching from `tokenizer.backend_tokenizer.pre_tokenizer` to `tokenizer.pre_tokenizer` in `tokenization_utils_tokenizers.py`; 45317 is the more complete version because it also adds tests.",
+        "confidence": 0.99,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "45317 is the better canonical PR because it implements the same bug fix as 45086 and includes accompanying test coverage, making it the more complete representative of the change.",
+        "best_issue_reason": null,
+        "best_pr_reason": "45317 is the best PR candidate because it addresses the same concrete Mistral pre-tokenizer AttributeError and adds validation tests, so it is the strongest standalone fix in the cluster.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:45086",
+            "right": "pull_request:45317",
+            "accept": true,
+            "reason": "Same underlying code-path bug in `_patch_mistral_regex`, same file, same explicit issue target, and the diffs overlap on the same fix; 45317 just adds tests."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 19,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "llm",
+      "cache_hit": false,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-45115-2",
+      "nodes": [
+        "pull_request:45115",
+        "pull_request:45569"
+      ],
+      "soft_pairs": [
+        "pull_request:45115|pull_request:45569"
+      ],
+      "prepared_review_unit_hash": "e6750721eb8783fdf126499ec017eeb37c80aed37bd362e5012beeec4196f714",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 6066,
+        "estimated_input_tokens": 1517,
+        "estimated_eval_tokens": 3290
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 6066,
+        "estimated_input_tokens": 1517,
+        "estimated_eval_tokens": 3290
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "The two PRs overlap on NemotronH codepaths, but they are not the same change: #45115 is a narrower inheritance/conversion-mapping refactor toward GraniteMoeHybrid compatibility, while #45569 is a broader model-family split introducing NemotronHDense/NemotronHSparse plus docs and auto-registration updates.",
+        "confidence": 0.92,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "#45569 is the more complete and user-facing PR: it defines the new NemotronH dense/sparse variants, updates docs, and wires the model into auto/config/model registration. It better represents the main architectural change in this cluster.",
+        "best_issue_reason": null,
+        "best_pr_reason": "#45569 is the best PR because it covers the full NemotronH refactor/splitting effort rather than only a conversion-mapping/inheritance adjustment.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:45115",
+            "right": "pull_request:45569",
+            "accept": false,
+            "reason": "They share some NemotronH files, but the underlying changes differ: #45115 focuses on checkpoint conversion naming and GraniteMoeHybrid-style inheritance, whereas #45569 introduces separate dense/sparse model variants and documentation/auto-map updates. They are related, not duplicates."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 20,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "llm",
+      "cache_hit": false,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-45332-2",
+      "nodes": [
+        "pull_request:45332",
+        "pull_request:45333"
+      ],
+      "soft_pairs": [
+        "pull_request:45332|pull_request:45333"
+      ],
+      "prepared_review_unit_hash": "c252bdbcbd96340711e2188c1c881d12f27b95eb3952546a91bf31e460b86297",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 5398,
+        "estimated_input_tokens": 1350,
+        "estimated_eval_tokens": 2956
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 5398,
+        "estimated_input_tokens": 1350,
+        "estimated_eval_tokens": 2956
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "Two open PRs propose related heterogeneous configuration/modeling work, but they are not duplicates: one adds per-layer config support, the other adds broader heterogeneous model support that builds on it. The larger PR is the better cluster representative.",
+        "confidence": 0.89,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "PR #45332 is the broader, more central change: it covers per-layer config plus the modeling/cache/masking updates that use it. PR #45333 is narrower and largely foundational/config-only.",
+        "best_issue_reason": null,
+        "best_pr_reason": "PR #45332 is the best representative of the cluster because it subsumes the shared heterogeneity groundwork and extends into the main runtime/modeling path, making it the more complete canonical PR.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:45332",
+            "right": "pull_request:45333",
+            "accept": false,
+            "reason": "Related feature work, but not the same concrete change or bug fix. #45333 focuses on per-layer configuration support, while #45332 adds broader heterogeneous model/runtime support that consumes that config. They are complementary rather than duplicate PRs."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 21,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "llm",
+      "cache_hit": false,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-45415-2",
+      "nodes": [
+        "pull_request:45415",
+        "pull_request:45425"
+      ],
+      "soft_pairs": [
+        "pull_request:45415|pull_request:45425"
+      ],
+      "prepared_review_unit_hash": "16137db7fa18870577d38ab41de5da181f80b0decfeaa00210acd97424fae61b",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 5514,
+        "estimated_input_tokens": 1379,
+        "estimated_eval_tokens": 3014
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 5514,
+        "estimated_input_tokens": 1379,
+        "estimated_eval_tokens": 3014
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "Two typing-focused PRs overlap in a few files, but they address different scopes: one is a broad type-checking sweep across many transformers modules, and the other adds specific typing protocols/helpers for modeling and PEFT-related code paths.",
+        "confidence": 0.82,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "PR #45425 is the more focused and concrete change: it adds shared typing aliases/protocols and updates a small set of code paths to use them. PR #45415 is a broader repository-wide typing cleanup/configuration pass, so it is less representative as the cluster anchor.",
+        "best_issue_reason": null,
+        "best_pr_reason": "PR #45425 is the best representative for this cluster because it centers on a specific typing abstraction change rather than a wide, multi-file type-checking sweep. PR #45415 looks like a larger housekeeping effort with many unrelated edits, making it a weaker canonical representative.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:45415",
+            "right": "pull_request:45425",
+            "accept": false,
+            "reason": "Do not merge: the overlap is only superficial typing-related cleanup. #45415 is a broad type-checking pass across many files and pyproject settings, while #45425 introduces specific typing protocols and helpers for modeling/PEFT code paths. They do not appear to fix the same concrete bug or change."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 22,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "llm",
+      "cache_hit": false,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-45487-4",
+      "nodes": [
+        "pull_request:45487",
+        "pull_request:45817",
+        "pull_request:45836",
+        "pull_request:45843"
+      ],
+      "soft_pairs": [
+        "pull_request:45487|pull_request:45843",
+        "pull_request:45487|pull_request:45836",
+        "pull_request:45817|pull_request:45843",
+        "pull_request:45836|pull_request:45843"
+      ],
+      "prepared_review_unit_hash": "6a96123a7dd219e95f907ba16cbb534a26673e315d9e8a31962abce02616020b",
+      "component_budget": {
+        "node_count": 4,
+        "item_count": 4,
+        "soft_pair_count": 4,
+        "serialized_chars": 10388,
+        "estimated_input_tokens": 2597,
+        "estimated_eval_tokens": 5450
+      },
+      "budget": {
+        "node_count": 4,
+        "item_count": 4,
+        "soft_pair_count": 4,
+        "serialized_chars": 10388,
+        "estimated_input_tokens": 2597,
+        "estimated_eval_tokens": 5450
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "This cluster is mostly a set of unrelated model-parallel/device-placement fixes across different model families, plus one unrelated test-skip PR. They are thematically similar but not the same underlying change.",
+        "confidence": 0.91,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "pull_request:45487 is the broadest and clearest representative of the shared model-parallel/device-placement theme, with concrete code changes rather than only test adjustments.",
+        "best_issue_reason": null,
+        "best_pr_reason": "pull_request:45487 best captures the cluster\u2019s main pattern of fixing model-parallel placement issues in model code; the others are either narrower, model-specific variants or unrelated test skips.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:45487",
+            "right": "pull_request:45843",
+            "accept": false,
+            "reason": "Both address model-parallel placement, but they fix different model families and different code paths; not the same concrete bug or mergeable as one PR."
+          },
+          {
+            "left": "pull_request:45487",
+            "right": "pull_request:45836",
+            "accept": false,
+            "reason": "45487 fixes runtime device-placement/model-parallel behavior, while 45836 only skips failing tests for a cache incompatibility; unrelated underlying change."
+          },
+          {
+            "left": "pull_request:45817",
+            "right": "pull_request:45843",
+            "accept": false,
+            "reason": "Similar high-level theme, but Gemma4 and Deimv2 have distinct implementations and fixes; not the same underlying bug."
+          },
+          {
+            "left": "pull_request:45836",
+            "right": "pull_request:45843",
+            "accept": false,
+            "reason": "Test-only skip for MiniCPM/Qwen cache incompatibility versus a Deimv2 model-parallel code fix; clearly different issues."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 23,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "llm",
+      "cache_hit": false,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-45548-2",
+      "nodes": [
+        "pull_request:45548",
+        "pull_request:45662"
+      ],
+      "soft_pairs": [
+        "pull_request:45548|pull_request:45662"
+      ],
+      "prepared_review_unit_hash": "70c5056f3f907c8ae594b89749e73943c9328affc7a2cee66a9e02228aa970c6",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 4505,
+        "estimated_input_tokens": 1127,
+        "estimated_eval_tokens": 2510
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 4505,
+        "estimated_input_tokens": 1127,
+        "estimated_eval_tokens": 2510
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "These two PRs both touch expert-parallel loading, but they address different bugs and code paths: one fixes EP + ZeRO-3/accelerate launch config handling, the other fixes EP + FSDP2 expert tensors being overwritten by rank-0 broadcast.",
+        "confidence": 0.94,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "No clear canonical duplicate here; PR 45662 is the more self-contained fix for a specific overwrite bug in distributed loading, while 45548 targets a separate ZeRO-3/config serialization issue.",
+        "best_issue_reason": null,
+        "best_pr_reason": "PR 45662 is the stronger representative of a concrete, narrowly scoped bug fix with multiple code-path changes, but it should not be treated as a duplicate of 45548.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:45548",
+            "right": "pull_request:45662",
+            "accept": false,
+            "reason": "Shared EP/distributed-loading theme is too broad; 45548 fixes accelerate/ZeRO-3 config handling and `distributed_config` serialization, while 45662 fixes FSDP2 rank-0 broadcast overwriting experts via `post_shard_wrap` and related loading paths."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 24,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "llm",
+      "cache_hit": false,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-45639-2",
+      "nodes": [
+        "pull_request:45639",
+        "pull_request:45645"
+      ],
+      "soft_pairs": [
+        "pull_request:45639|pull_request:45645"
+      ],
+      "prepared_review_unit_hash": "7bdffe171c433d63389d9812273698bf553b4f96b3aa95218a39f09d39882fc2",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 4589,
+        "estimated_input_tokens": 1148,
+        "estimated_eval_tokens": 2552
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 4589,
+        "estimated_input_tokens": 1148,
+        "estimated_eval_tokens": 2552
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "Both PRs target the same xdist-related captured_info logging problem; PR 45645 appears to subsume PR 45639 with the same testing_utils fix plus the CI artifact handling changes.",
+        "confidence": 0.89,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "PR 45645 is the broader, more complete fix for the captured_info xdist collision issue: it includes the testing_utils change plus workflow/notification updates to handle multiple captured_info files in CI.",
+        "best_issue_reason": null,
+        "best_pr_reason": "PR 45645 is the best representative because it addresses the underlying bug end-to-end, while 45639 is a narrower subset limited to making the debug log write path xdist-safe.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:45639",
+            "right": "pull_request:45645",
+            "accept": true,
+            "reason": "Same underlying xdist/captured_info collision problem, same issue target, and 45645 appears to extend rather than diverge from 45639; these could plausibly be merged into one PR."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 25,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-45683-2",
+      "nodes": [
+        "pull_request:45683",
+        "pull_request:45734"
+      ],
+      "soft_pairs": [
+        "pull_request:45683|pull_request:45734"
+      ],
+      "prepared_review_unit_hash": "eba4ca0f26b2d3be0c00e316270a9341d31af2208b44dbab31b1d3a8e81e7538",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 4469,
+        "estimated_input_tokens": 1118,
+        "estimated_eval_tokens": 2492
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 4469,
+        "estimated_input_tokens": 1118,
+        "estimated_eval_tokens": 2492
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "Two open PRs touch the same quantization helper, but they fix different bugs: one special-cases audio submodules, the other changes how default skip lists are combined with user-supplied skips. They should not be clustered as duplicates.",
+        "confidence": 0.93,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "45734 is the better representative PR because it fixes the core skip-module composition logic in `get_modules_to_not_convert`, which is broader and more central than the narrow audio-module special case in 45683.",
+        "best_issue_reason": null,
+        "best_pr_reason": "45734 is the strongest standalone PR in the set: it addresses a concrete quantization failure mode in the main code path and its fix is directly about the helper\u2019s behavior rather than a model-specific workaround.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:45683",
+            "right": "pull_request:45734",
+            "accept": false,
+            "reason": "Both touch quantization skip logic, but 45683 adds audio-tower/embed_audio exclusions while 45734 changes additive handling of `skip_modules`; these are different underlying bugs and would not naturally merge into one PR."
+          }
+        ]
+      },
+      "evaluator_result": null,
+      "evaluator_used": false,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": false,
+      "error_kind": null,
+      "error_message": null
+    },
+    {
+      "label": "pull_request",
+      "component_index": 26,
+      "component_count": 26,
+      "review_unit_index": 1,
+      "review_unit_count": 1,
+      "status": "reviewed",
+      "reason": null,
+      "source": "cache",
+      "cache_hit": true,
+      "model": "gpt-5.4-mini?service_tier=flex",
+      "cluster_id": "cluster-45737-2",
+      "nodes": [
+        "pull_request:45737",
+        "pull_request:45738"
+      ],
+      "soft_pairs": [
+        "pull_request:45737|pull_request:45738"
+      ],
+      "prepared_review_unit_hash": "cc2ade73a49ec6a8883a7a24c433988645588bcd049e8aad2bce56b1b24daa1c",
+      "component_budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 4131,
+        "estimated_input_tokens": 1033,
+        "estimated_eval_tokens": 2322
+      },
+      "budget": {
+        "node_count": 2,
+        "item_count": 2,
+        "soft_pair_count": 1,
+        "serialized_chars": 4131,
+        "estimated_input_tokens": 1033,
+        "estimated_eval_tokens": 2322
+      },
+      "overflow_policy": "truncate_then_skip",
+      "trimmed": false,
+      "aggressively_trimmed": false,
+      "split": false,
+      "analyst_result": {
+        "summary": "Both PRs address the same MusicgenMelody cache initialization bug in `forward()`: replacing `EncoderDecoderCache(DynamicCache, DynamicCache)` with a single `DynamicCache`. PR #45737 is the more complete duplicate because it bundles the fix with a regression test, while #45738 is the fix-only variant.",
+        "confidence": 0.97,
+        "canonical_issue_reason": null,
+        "canonical_pr_reason": "PR #45737 is the best canonical PR because it contains the same code fix as #45738 plus a regression test, making it the most complete representation of the bug fix.",
+        "best_issue_reason": null,
+        "best_pr_reason": "PR #45737 is the strongest overall candidate: it fixes the concrete cache bug and adds a targeted regression test, whereas #45738 only changes the implementation.",
+        "soft_edge_verdicts": [
+          {
+            "left": "pull_request:45737",
+            "right": "pull_request:45738",
+            "accept": true,
+            "reason": "Same underlying code-path bug in `musicgen_melody` cache setup; the diffs match on the core fix and the test-only addition in #45737 could plausibly live in the same PR as #45738."
+          }
+        ]
+      },
+      "evaluator_result": {
+        "accept": true,
+        "feedback": "The assessment is grounded in the packet: both PRs make the same `musicgen_melody` cache initialization change, and #45737 only adds a regression test on top of that fix. The soft-edge accept is conservative enough because the diffs are mergeable into one PR for the same bugfix."
+      },
+      "evaluator_used": true,
+      "retried": false,
+      "accepted_nontrivial_soft_edge": true,
+      "error_kind": null,
+      "error_message": null
+    }
+  ]
+}
diff --git a/snapshots/20260510T060019Z/analysis-runs/hybrid-model-20260510t060019z/manifest.json b/snapshots/20260510T060019Z/analysis-runs/hybrid-model-20260510t060019z/manifest.json
new file mode 100644
index 0000000000000000000000000000000000000000..b9c5532f7682181a08c4f08262ad0b1fe1f5a270
--- /dev/null
+++ b/snapshots/20260510T060019Z/analysis-runs/hybrid-model-20260510t060019z/manifest.json
@@ -0,0 +1,14 @@
+{
+  "analysis_id": "hybrid-model-20260510t060019z",
+  "artifacts": {
+    "hybrid": "snapshots/20260510T060019Z/analysis-runs/hybrid-model-20260510t060019z/analysis-report-hybrid.json",
+    "hybrid_reviews": "snapshots/20260510T060019Z/analysis-runs/hybrid-model-20260510t060019z/analysis-report-hybrid.llm-reviews.json"
+  },
+  "channel": "canonical",
+  "model": null,
+  "published_at": "2026-05-10T06:03:19Z",
+  "repo": "huggingface/transformers",
+  "schema_version": 1,
+  "snapshot_id": "20260510T060019Z",
+  "variant": "hybrid"
+}
diff --git a/snapshots/20260510T060019Z/manifest.json b/snapshots/20260510T060019Z/manifest.json
index ead106d3147b1a0bb8e5e082293a292248f0d68b..83e1e06e3d90b4fced569e08e575ae3c7590ee3d 100644
--- a/snapshots/20260510T060019Z/manifest.json
+++ b/snapshots/20260510T060019Z/manifest.json
@@ -32,6 +32,24 @@
     "timeline_events": 50
   },
   "extracted_at": "2026-05-10T06:00:19Z",
+  "published_analysis": {
+    "canonical_analysis_id": "hybrid-model-20260510t060019z",
+    "runs": {
+      "hybrid-model-20260510t060019z": {
+        "analysis_id": "hybrid-model-20260510t060019z",
+        "artifacts": {
+          "hybrid": "snapshots/20260510T060019Z/analysis-runs/hybrid-model-20260510t060019z/analysis-report-hybrid.json",
+          "hybrid_reviews": "snapshots/20260510T060019Z/analysis-runs/hybrid-model-20260510t060019z/analysis-report-hybrid.llm-reviews.json"
+        },
+        "channel": "canonical",
+        "manifest_path": "snapshots/20260510T060019Z/analysis-runs/hybrid-model-20260510t060019z/manifest.json",
+        "model": null,
+        "published_at": "2026-05-10T06:03:19Z",
+        "variant": "hybrid"
+      }
+    },
+    "schema_version": 1
+  },
   "repo": "huggingface/transformers",
   "snapshot_id": "20260510T060019Z",
   "watermark": {