ConorWang commited on
Commit
0a1c0ab
·
verified ·
1 Parent(s): 6311fba

Upload rollback adapter artifacts to rollback_adapter/

Browse files
.gitattributes CHANGED
@@ -37,3 +37,4 @@ tokenizer.json filter=lfs diff=lfs merge=lfs -text
37
  toolspec_adapter/tokenizer/tokenizer.json filter=lfs diff=lfs merge=lfs -text
38
  uncertainty_adapter/tokenizer/tokenizer.json filter=lfs diff=lfs merge=lfs -text
39
  evidence_adapter/tokenizer/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 
 
37
  toolspec_adapter/tokenizer/tokenizer.json filter=lfs diff=lfs merge=lfs -text
38
  uncertainty_adapter/tokenizer/tokenizer.json filter=lfs diff=lfs merge=lfs -text
39
  evidence_adapter/tokenizer/tokenizer.json filter=lfs diff=lfs merge=lfs -text
40
+ rollback_adapter/tokenizer/tokenizer.json filter=lfs diff=lfs merge=lfs -text
rollback_adapter/adapter/README.md ADDED
@@ -0,0 +1,203 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: peft
3
+ tags:
4
+ - lora
5
+ ---
6
+
7
+ # Model Card for Model ID
8
+
9
+ <!-- Provide a quick summary of what the model is/does. -->
10
+
11
+
12
+
13
+ ## Model Details
14
+
15
+ ### Model Description
16
+
17
+ <!-- Provide a longer summary of what this model is. -->
18
+
19
+
20
+
21
+ - **Developed by:** [More Information Needed]
22
+ - **Funded by [optional]:** [More Information Needed]
23
+ - **Shared by [optional]:** [More Information Needed]
24
+ - **Model type:** [More Information Needed]
25
+ - **Language(s) (NLP):** [More Information Needed]
26
+ - **License:** [More Information Needed]
27
+ - **Finetuned from model [optional]:** [More Information Needed]
28
+
29
+ ### Model Sources [optional]
30
+
31
+ <!-- Provide the basic links for the model. -->
32
+
33
+ - **Repository:** [More Information Needed]
34
+ - **Paper [optional]:** [More Information Needed]
35
+ - **Demo [optional]:** [More Information Needed]
36
+
37
+ ## Uses
38
+
39
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
40
+
41
+ ### Direct Use
42
+
43
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
44
+
45
+ [More Information Needed]
46
+
47
+ ### Downstream Use [optional]
48
+
49
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
50
+
51
+ [More Information Needed]
52
+
53
+ ### Out-of-Scope Use
54
+
55
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
56
+
57
+ [More Information Needed]
58
+
59
+ ## Bias, Risks, and Limitations
60
+
61
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
62
+
63
+ [More Information Needed]
64
+
65
+ ### Recommendations
66
+
67
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
68
+
69
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
70
+
71
+ ## How to Get Started with the Model
72
+
73
+ Use the code below to get started with the model.
74
+
75
+ [More Information Needed]
76
+
77
+ ## Training Details
78
+
79
+ ### Training Data
80
+
81
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
82
+
83
+ [More Information Needed]
84
+
85
+ ### Training Procedure
86
+
87
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
88
+
89
+ #### Preprocessing [optional]
90
+
91
+ [More Information Needed]
92
+
93
+
94
+ #### Training Hyperparameters
95
+
96
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
97
+
98
+ #### Speeds, Sizes, Times [optional]
99
+
100
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
101
+
102
+ [More Information Needed]
103
+
104
+ ## Evaluation
105
+
106
+ <!-- This section describes the evaluation protocols and provides the results. -->
107
+
108
+ ### Testing Data, Factors & Metrics
109
+
110
+ #### Testing Data
111
+
112
+ <!-- This should link to a Dataset Card if possible. -->
113
+
114
+ [More Information Needed]
115
+
116
+ #### Factors
117
+
118
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
119
+
120
+ [More Information Needed]
121
+
122
+ #### Metrics
123
+
124
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
125
+
126
+ [More Information Needed]
127
+
128
+ ### Results
129
+
130
+ [More Information Needed]
131
+
132
+ #### Summary
133
+
134
+
135
+
136
+ ## Model Examination [optional]
137
+
138
+ <!-- Relevant interpretability work for the model goes here -->
139
+
140
+ [More Information Needed]
141
+
142
+ ## Environmental Impact
143
+
144
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
145
+
146
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
147
+
148
+ - **Hardware Type:** [More Information Needed]
149
+ - **Hours used:** [More Information Needed]
150
+ - **Cloud Provider:** [More Information Needed]
151
+ - **Compute Region:** [More Information Needed]
152
+ - **Carbon Emitted:** [More Information Needed]
153
+
154
+ ## Technical Specifications [optional]
155
+
156
+ ### Model Architecture and Objective
157
+
158
+ [More Information Needed]
159
+
160
+ ### Compute Infrastructure
161
+
162
+ [More Information Needed]
163
+
164
+ #### Hardware
165
+
166
+ [More Information Needed]
167
+
168
+ #### Software
169
+
170
+ [More Information Needed]
171
+
172
+ ## Citation [optional]
173
+
174
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
175
+
176
+ **BibTeX:**
177
+
178
+ [More Information Needed]
179
+
180
+ **APA:**
181
+
182
+ [More Information Needed]
183
+
184
+ ## Glossary [optional]
185
+
186
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
187
+
188
+ [More Information Needed]
189
+
190
+ ## More Information [optional]
191
+
192
+ [More Information Needed]
193
+
194
+ ## Model Card Authors [optional]
195
+
196
+ [More Information Needed]
197
+
198
+ ## Model Card Contact
199
+
200
+ [More Information Needed]
201
+ ### Framework versions
202
+
203
+ - PEFT 0.19.0
rollback_adapter/adapter/adapter_config.json ADDED
@@ -0,0 +1,57 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": null,
7
+ "bias": "none",
8
+ "corda_config": null,
9
+ "ensure_weight_tying": false,
10
+ "eva_config": null,
11
+ "exclude_modules": null,
12
+ "fan_in_fan_out": false,
13
+ "inference_mode": true,
14
+ "init_lora_weights": true,
15
+ "layer_replication": null,
16
+ "layers_pattern": null,
17
+ "layers_to_transform": null,
18
+ "loftq_config": {},
19
+ "lora_alpha": 16,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.05,
22
+ "lora_ga_config": null,
23
+ "megatron_config": null,
24
+ "megatron_core": "megatron.core",
25
+ "modules_to_save": null,
26
+ "peft_type": "LORA",
27
+ "peft_version": "0.19.0",
28
+ "qalora_group_size": 16,
29
+ "r": 8,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "surface_host.rollback_adapter.head",
34
+ "surface_host.toolspec_head.receipt_formatter",
35
+ "surface_host.runtime_harness_adapter.bridge",
36
+ "surface_host.toolspec_bridge.adapter",
37
+ "surface_host.failure_signal_bridge.rollback_bridge",
38
+ "surface_host.sandbox_rollback_bridge.adapter",
39
+ "surface_host.validator_feedback_bridge.adapter",
40
+ "surface_host.validator_feedback_loop.rollback_adapter",
41
+ "surface_host.request_normalizer.adapter",
42
+ "surface_host.tool_protocol_adapter.bridge",
43
+ "surface_host.toolspec_head.precondition_adapter",
44
+ "surface_host.memory_boundary_guard.adapter",
45
+ "surface_host.toolspec_head.postcondition_adapter",
46
+ "surface_host.memory_boundary_guard.rollback_filter",
47
+ "surface_host.toolspec_head.validator_gate",
48
+ "surface_host.toolspec_head.trigger_gate"
49
+ ],
50
+ "target_parameters": null,
51
+ "task_type": "FEATURE_EXTRACTION",
52
+ "trainable_token_indices": null,
53
+ "use_bdlora": null,
54
+ "use_dora": false,
55
+ "use_qalora": false,
56
+ "use_rslora": false
57
+ }
rollback_adapter/adapter/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:74ca61f2f73640249123de75d35d15493a95e1a70c8ed37db4b0cacf668046b7
3
+ size 1053176
rollback_adapter/best_checkpoint_manifest.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_epoch": 3,
3
+ "best_quality_score": 0.3986773560841878,
4
+ "eval_metrics": {
5
+ "avg_binary_accuracy": 0.8194444444444444,
6
+ "cause_accuracy": 0.4888888888888889,
7
+ "count": 45,
8
+ "eval_batches": 45,
9
+ "eval_loss": 7.656409973568387,
10
+ "fail_closed_accuracy": 0.7777777777777778,
11
+ "minimal_diff_accuracy": 0.5555555555555556,
12
+ "mode_accuracy": 0.3333333333333333,
13
+ "need_accuracy": 0.7777777777777778,
14
+ "preserve_conflict_accuracy": 0.7777777777777778,
15
+ "quality_score": 0.3986773560841878,
16
+ "reverse_engineering_bounded_accuracy": 0.8888888888888888,
17
+ "rewrite_accuracy": 0.4888888888888889,
18
+ "selfcheck_route_accuracy": 0.5555555555555556,
19
+ "session_continuity_accuracy": 1.0,
20
+ "validator_aligned_accuracy": 1.0,
21
+ "worktree_safe_accuracy": 1.0
22
+ },
23
+ "train_metrics": {
24
+ "loss": 0.4452953128114579,
25
+ "micro_batches": 126,
26
+ "optimizer_steps": 8
27
+ }
28
+ }
rollback_adapter/epoch_history.json ADDED
@@ -0,0 +1,120 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epochs": [
3
+ {
4
+ "epoch": 1,
5
+ "eval_metrics": {
6
+ "avg_binary_accuracy": 0.8194444444444444,
7
+ "cause_accuracy": 0.4444444444444444,
8
+ "count": 45,
9
+ "eval_batches": 45,
10
+ "eval_loss": 8.320836893717448,
11
+ "fail_closed_accuracy": 0.7777777777777778,
12
+ "minimal_diff_accuracy": 0.5555555555555556,
13
+ "mode_accuracy": 0.26666666666666666,
14
+ "need_accuracy": 0.7777777777777778,
15
+ "preserve_conflict_accuracy": 0.7777777777777778,
16
+ "quality_score": 0.30316659545898433,
17
+ "reverse_engineering_bounded_accuracy": 0.8888888888888888,
18
+ "rewrite_accuracy": 0.28888888888888886,
19
+ "selfcheck_route_accuracy": 0.5555555555555556,
20
+ "session_continuity_accuracy": 1.0,
21
+ "validator_aligned_accuracy": 1.0,
22
+ "worktree_safe_accuracy": 1.0
23
+ },
24
+ "improved": true,
25
+ "quality_score": 0.30316659545898433,
26
+ "train_metrics": {
27
+ "loss": 0.5652040739854177,
28
+ "micro_batches": 126,
29
+ "optimizer_steps": 8
30
+ }
31
+ },
32
+ {
33
+ "epoch": 2,
34
+ "eval_metrics": {
35
+ "avg_binary_accuracy": 0.8194444444444444,
36
+ "cause_accuracy": 0.3333333333333333,
37
+ "count": 45,
38
+ "eval_batches": 45,
39
+ "eval_loss": 7.85913037194146,
40
+ "fail_closed_accuracy": 0.7777777777777778,
41
+ "minimal_diff_accuracy": 0.5555555555555556,
42
+ "mode_accuracy": 0.4,
43
+ "need_accuracy": 0.7777777777777778,
44
+ "preserve_conflict_accuracy": 0.7777777777777778,
45
+ "quality_score": 0.28684517033894863,
46
+ "reverse_engineering_bounded_accuracy": 0.8888888888888888,
47
+ "rewrite_accuracy": 0.28888888888888886,
48
+ "selfcheck_route_accuracy": 0.5555555555555556,
49
+ "session_continuity_accuracy": 1.0,
50
+ "validator_aligned_accuracy": 1.0,
51
+ "worktree_safe_accuracy": 1.0
52
+ },
53
+ "improved": false,
54
+ "quality_score": 0.28684517033894863,
55
+ "train_metrics": {
56
+ "loss": 0.4876745880123169,
57
+ "micro_batches": 126,
58
+ "optimizer_steps": 8
59
+ }
60
+ },
61
+ {
62
+ "epoch": 3,
63
+ "eval_metrics": {
64
+ "avg_binary_accuracy": 0.8194444444444444,
65
+ "cause_accuracy": 0.4888888888888889,
66
+ "count": 45,
67
+ "eval_batches": 45,
68
+ "eval_loss": 7.656409973568387,
69
+ "fail_closed_accuracy": 0.7777777777777778,
70
+ "minimal_diff_accuracy": 0.5555555555555556,
71
+ "mode_accuracy": 0.3333333333333333,
72
+ "need_accuracy": 0.7777777777777778,
73
+ "preserve_conflict_accuracy": 0.7777777777777778,
74
+ "quality_score": 0.3986773560841878,
75
+ "reverse_engineering_bounded_accuracy": 0.8888888888888888,
76
+ "rewrite_accuracy": 0.4888888888888889,
77
+ "selfcheck_route_accuracy": 0.5555555555555556,
78
+ "session_continuity_accuracy": 1.0,
79
+ "validator_aligned_accuracy": 1.0,
80
+ "worktree_safe_accuracy": 1.0
81
+ },
82
+ "improved": true,
83
+ "quality_score": 0.3986773560841878,
84
+ "train_metrics": {
85
+ "loss": 0.4452953128114579,
86
+ "micro_batches": 126,
87
+ "optimizer_steps": 8
88
+ }
89
+ },
90
+ {
91
+ "epoch": 4,
92
+ "eval_metrics": {
93
+ "avg_binary_accuracy": 0.8194444444444444,
94
+ "cause_accuracy": 0.4888888888888889,
95
+ "count": 45,
96
+ "eval_batches": 45,
97
+ "eval_loss": 7.64718263414171,
98
+ "fail_closed_accuracy": 0.7777777777777778,
99
+ "minimal_diff_accuracy": 0.5555555555555556,
100
+ "mode_accuracy": 0.3333333333333333,
101
+ "need_accuracy": 0.7777777777777778,
102
+ "preserve_conflict_accuracy": 0.7777777777777778,
103
+ "quality_score": 0.3988619028727214,
104
+ "reverse_engineering_bounded_accuracy": 0.8888888888888888,
105
+ "rewrite_accuracy": 0.4888888888888889,
106
+ "selfcheck_route_accuracy": 0.5555555555555556,
107
+ "session_continuity_accuracy": 1.0,
108
+ "validator_aligned_accuracy": 1.0,
109
+ "worktree_safe_accuracy": 1.0
110
+ },
111
+ "improved": false,
112
+ "quality_score": 0.3988619028727214,
113
+ "train_metrics": {
114
+ "loss": 0.4358076933357451,
115
+ "micro_batches": 126,
116
+ "optimizer_steps": 8
117
+ }
118
+ }
119
+ ]
120
+ }
rollback_adapter/host_manifest.json ADDED
@@ -0,0 +1,288 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hidden_size": 2048,
3
+ "host_config": {
4
+ "attn_implementation": null,
5
+ "backbone_name_or_path": "/public/wang_libo/veriloop_coder_e1/model",
6
+ "device_map": null,
7
+ "dtype": null,
8
+ "evidence_rank_hint": 8,
9
+ "expose_backbone_inventory": false,
10
+ "freeze_backbone": true,
11
+ "hidden_size_override": 2048,
12
+ "host_dropout": 0.0,
13
+ "identity_rank_hint": 8,
14
+ "load_backbone_weights": false,
15
+ "local_files_only": true,
16
+ "low_cpu_mem_usage": true,
17
+ "memory_rank_hint": 4,
18
+ "rollback_rank_hint": 8,
19
+ "runtime_rank_hint": 8,
20
+ "toolspec_rank_hint": 8,
21
+ "trust_remote_code": true,
22
+ "uncertainty_rank_hint": 8,
23
+ "use_safetensors": null,
24
+ "validator_rank_hint": 8
25
+ },
26
+ "load_result": {
27
+ "has_base_config": true,
28
+ "has_base_model": true,
29
+ "hidden_size": 2048,
30
+ "notes": [
31
+ "class=AutoModelForCausalLM",
32
+ "quant=4bit"
33
+ ],
34
+ "source": "trainer_qwen36_loader"
35
+ },
36
+ "peft_named_modules": [
37
+ "claim_extractor",
38
+ "claim_extractor.adapter",
39
+ "dropout",
40
+ "episodic_memory",
41
+ "episodic_memory.adapter",
42
+ "evidence_binding",
43
+ "evidence_binding.adapter",
44
+ "failure_signal_bridge",
45
+ "failure_signal_bridge.rollback_bridge",
46
+ "failure_signal_bridge.rollback_bridge.base_layer",
47
+ "failure_signal_bridge.rollback_bridge.lora_A",
48
+ "failure_signal_bridge.rollback_bridge.lora_A.default",
49
+ "failure_signal_bridge.rollback_bridge.lora_B",
50
+ "failure_signal_bridge.rollback_bridge.lora_B.default",
51
+ "failure_signal_bridge.rollback_bridge.lora_dropout",
52
+ "failure_signal_bridge.rollback_bridge.lora_dropout.default",
53
+ "failure_signal_bridge.rollback_bridge.lora_embedding_A",
54
+ "failure_signal_bridge.rollback_bridge.lora_embedding_B",
55
+ "failure_signal_bridge.rollback_bridge.lora_magnitude_vector",
56
+ "identity_adapter",
57
+ "identity_adapter.bridge",
58
+ "identity_guard",
59
+ "identity_guard.adapter",
60
+ "input_norm",
61
+ "memory_boundary_guard",
62
+ "memory_boundary_guard.adapter",
63
+ "memory_boundary_guard.adapter.base_layer",
64
+ "memory_boundary_guard.adapter.lora_A",
65
+ "memory_boundary_guard.adapter.lora_A.default",
66
+ "memory_boundary_guard.adapter.lora_B",
67
+ "memory_boundary_guard.adapter.lora_B.default",
68
+ "memory_boundary_guard.adapter.lora_dropout",
69
+ "memory_boundary_guard.adapter.lora_dropout.default",
70
+ "memory_boundary_guard.adapter.lora_embedding_A",
71
+ "memory_boundary_guard.adapter.lora_embedding_B",
72
+ "memory_boundary_guard.adapter.lora_magnitude_vector",
73
+ "memory_boundary_guard.rollback_filter",
74
+ "memory_boundary_guard.rollback_filter.base_layer",
75
+ "memory_boundary_guard.rollback_filter.lora_A",
76
+ "memory_boundary_guard.rollback_filter.lora_A.default",
77
+ "memory_boundary_guard.rollback_filter.lora_B",
78
+ "memory_boundary_guard.rollback_filter.lora_B.default",
79
+ "memory_boundary_guard.rollback_filter.lora_dropout",
80
+ "memory_boundary_guard.rollback_filter.lora_dropout.default",
81
+ "memory_boundary_guard.rollback_filter.lora_embedding_A",
82
+ "memory_boundary_guard.rollback_filter.lora_embedding_B",
83
+ "memory_boundary_guard.rollback_filter.lora_magnitude_vector",
84
+ "patch_binding",
85
+ "patch_binding.adapter",
86
+ "permission_context_manager",
87
+ "permission_context_manager.adapter",
88
+ "progress_state_tracker",
89
+ "progress_state_tracker.adapter",
90
+ "progress_state_tracker.rollback_memory",
91
+ "proof_carrying_hints",
92
+ "proof_carrying_hints.bridge",
93
+ "provenance_binding",
94
+ "provenance_binding.adapter",
95
+ "public_identity_head",
96
+ "public_identity_head.proj",
97
+ "query_runtime_engine",
98
+ "query_runtime_engine.adapter",
99
+ "request_normalizer",
100
+ "request_normalizer.adapter",
101
+ "request_normalizer.adapter.base_layer",
102
+ "request_normalizer.adapter.lora_A",
103
+ "request_normalizer.adapter.lora_A.default",
104
+ "request_normalizer.adapter.lora_B",
105
+ "request_normalizer.adapter.lora_B.default",
106
+ "request_normalizer.adapter.lora_dropout",
107
+ "request_normalizer.adapter.lora_dropout.default",
108
+ "request_normalizer.adapter.lora_embedding_A",
109
+ "request_normalizer.adapter.lora_embedding_B",
110
+ "request_normalizer.adapter.lora_magnitude_vector",
111
+ "rollback_adapter",
112
+ "rollback_adapter.head",
113
+ "rollback_adapter.head.base_layer",
114
+ "rollback_adapter.head.lora_A",
115
+ "rollback_adapter.head.lora_A.default",
116
+ "rollback_adapter.head.lora_B",
117
+ "rollback_adapter.head.lora_B.default",
118
+ "rollback_adapter.head.lora_dropout",
119
+ "rollback_adapter.head.lora_dropout.default",
120
+ "rollback_adapter.head.lora_embedding_A",
121
+ "rollback_adapter.head.lora_embedding_B",
122
+ "rollback_adapter.head.lora_magnitude_vector",
123
+ "rollback_engine",
124
+ "rollback_engine.adapter",
125
+ "runtime_binding",
126
+ "runtime_binding.adapter",
127
+ "runtime_harness_adapter",
128
+ "runtime_harness_adapter.bridge",
129
+ "runtime_harness_adapter.bridge.base_layer",
130
+ "runtime_harness_adapter.bridge.lora_A",
131
+ "runtime_harness_adapter.bridge.lora_A.default",
132
+ "runtime_harness_adapter.bridge.lora_B",
133
+ "runtime_harness_adapter.bridge.lora_B.default",
134
+ "runtime_harness_adapter.bridge.lora_dropout",
135
+ "runtime_harness_adapter.bridge.lora_dropout.default",
136
+ "runtime_harness_adapter.bridge.lora_embedding_A",
137
+ "runtime_harness_adapter.bridge.lora_embedding_B",
138
+ "runtime_harness_adapter.bridge.lora_magnitude_vector",
139
+ "runtime_harness_uncertainty_bridge",
140
+ "runtime_harness_uncertainty_bridge.adapter",
141
+ "sandbox_rollback_bridge",
142
+ "sandbox_rollback_bridge.adapter",
143
+ "sandbox_rollback_bridge.adapter.base_layer",
144
+ "sandbox_rollback_bridge.adapter.lora_A",
145
+ "sandbox_rollback_bridge.adapter.lora_A.default",
146
+ "sandbox_rollback_bridge.adapter.lora_B",
147
+ "sandbox_rollback_bridge.adapter.lora_B.default",
148
+ "sandbox_rollback_bridge.adapter.lora_dropout",
149
+ "sandbox_rollback_bridge.adapter.lora_dropout.default",
150
+ "sandbox_rollback_bridge.adapter.lora_embedding_A",
151
+ "sandbox_rollback_bridge.adapter.lora_embedding_B",
152
+ "sandbox_rollback_bridge.adapter.lora_magnitude_vector",
153
+ "session_compactor",
154
+ "session_compactor.adapter",
155
+ "session_state_manager",
156
+ "session_state_manager.adapter",
157
+ "session_state_manager.rollback_state",
158
+ "tool_protocol_adapter",
159
+ "tool_protocol_adapter.bridge",
160
+ "tool_protocol_adapter.bridge.base_layer",
161
+ "tool_protocol_adapter.bridge.lora_A",
162
+ "tool_protocol_adapter.bridge.lora_A.default",
163
+ "tool_protocol_adapter.bridge.lora_B",
164
+ "tool_protocol_adapter.bridge.lora_B.default",
165
+ "tool_protocol_adapter.bridge.lora_dropout",
166
+ "tool_protocol_adapter.bridge.lora_dropout.default",
167
+ "tool_protocol_adapter.bridge.lora_embedding_A",
168
+ "tool_protocol_adapter.bridge.lora_embedding_B",
169
+ "tool_protocol_adapter.bridge.lora_magnitude_vector",
170
+ "toolspec_bridge",
171
+ "toolspec_bridge.adapter",
172
+ "toolspec_bridge.adapter.base_layer",
173
+ "toolspec_bridge.adapter.lora_A",
174
+ "toolspec_bridge.adapter.lora_A.default",
175
+ "toolspec_bridge.adapter.lora_B",
176
+ "toolspec_bridge.adapter.lora_B.default",
177
+ "toolspec_bridge.adapter.lora_dropout",
178
+ "toolspec_bridge.adapter.lora_dropout.default",
179
+ "toolspec_bridge.adapter.lora_embedding_A",
180
+ "toolspec_bridge.adapter.lora_embedding_B",
181
+ "toolspec_bridge.adapter.lora_magnitude_vector",
182
+ "toolspec_head",
183
+ "toolspec_head.param_schema_adapter",
184
+ "toolspec_head.postcondition_adapter",
185
+ "toolspec_head.postcondition_adapter.base_layer",
186
+ "toolspec_head.postcondition_adapter.lora_A",
187
+ "toolspec_head.postcondition_adapter.lora_A.default",
188
+ "toolspec_head.postcondition_adapter.lora_B",
189
+ "toolspec_head.postcondition_adapter.lora_B.default",
190
+ "toolspec_head.postcondition_adapter.lora_dropout",
191
+ "toolspec_head.postcondition_adapter.lora_dropout.default",
192
+ "toolspec_head.postcondition_adapter.lora_embedding_A",
193
+ "toolspec_head.postcondition_adapter.lora_embedding_B",
194
+ "toolspec_head.postcondition_adapter.lora_magnitude_vector",
195
+ "toolspec_head.precondition_adapter",
196
+ "toolspec_head.precondition_adapter.base_layer",
197
+ "toolspec_head.precondition_adapter.lora_A",
198
+ "toolspec_head.precondition_adapter.lora_A.default",
199
+ "toolspec_head.precondition_adapter.lora_B",
200
+ "toolspec_head.precondition_adapter.lora_B.default",
201
+ "toolspec_head.precondition_adapter.lora_dropout",
202
+ "toolspec_head.precondition_adapter.lora_dropout.default",
203
+ "toolspec_head.precondition_adapter.lora_embedding_A",
204
+ "toolspec_head.precondition_adapter.lora_embedding_B",
205
+ "toolspec_head.precondition_adapter.lora_magnitude_vector",
206
+ "toolspec_head.receipt_formatter",
207
+ "toolspec_head.receipt_formatter.base_layer",
208
+ "toolspec_head.receipt_formatter.lora_A",
209
+ "toolspec_head.receipt_formatter.lora_A.default",
210
+ "toolspec_head.receipt_formatter.lora_B",
211
+ "toolspec_head.receipt_formatter.lora_B.default",
212
+ "toolspec_head.receipt_formatter.lora_dropout",
213
+ "toolspec_head.receipt_formatter.lora_dropout.default",
214
+ "toolspec_head.receipt_formatter.lora_embedding_A",
215
+ "toolspec_head.receipt_formatter.lora_embedding_B",
216
+ "toolspec_head.receipt_formatter.lora_magnitude_vector",
217
+ "toolspec_head.trigger_gate",
218
+ "toolspec_head.trigger_gate.base_layer",
219
+ "toolspec_head.trigger_gate.lora_A",
220
+ "toolspec_head.trigger_gate.lora_A.default",
221
+ "toolspec_head.trigger_gate.lora_B",
222
+ "toolspec_head.trigger_gate.lora_B.default",
223
+ "toolspec_head.trigger_gate.lora_dropout",
224
+ "toolspec_head.trigger_gate.lora_dropout.default",
225
+ "toolspec_head.trigger_gate.lora_embedding_A",
226
+ "toolspec_head.trigger_gate.lora_embedding_B",
227
+ "toolspec_head.trigger_gate.lora_magnitude_vector",
228
+ "toolspec_head.validator_gate",
229
+ "toolspec_head.validator_gate.base_layer",
230
+ "toolspec_head.validator_gate.lora_A",
231
+ "toolspec_head.validator_gate.lora_A.default",
232
+ "toolspec_head.validator_gate.lora_B",
233
+ "toolspec_head.validator_gate.lora_B.default",
234
+ "toolspec_head.validator_gate.lora_dropout",
235
+ "toolspec_head.validator_gate.lora_dropout.default",
236
+ "toolspec_head.validator_gate.lora_embedding_A",
237
+ "toolspec_head.validator_gate.lora_embedding_B",
238
+ "toolspec_head.validator_gate.lora_magnitude_vector",
239
+ "uncertainty_head",
240
+ "uncertainty_head.calibration_mlp",
241
+ "uncertainty_head.proj",
242
+ "validator_feedback_bridge",
243
+ "validator_feedback_bridge.adapter",
244
+ "validator_feedback_bridge.adapter.base_layer",
245
+ "validator_feedback_bridge.adapter.lora_A",
246
+ "validator_feedback_bridge.adapter.lora_A.default",
247
+ "validator_feedback_bridge.adapter.lora_B",
248
+ "validator_feedback_bridge.adapter.lora_B.default",
249
+ "validator_feedback_bridge.adapter.lora_dropout",
250
+ "validator_feedback_bridge.adapter.lora_dropout.default",
251
+ "validator_feedback_bridge.adapter.lora_embedding_A",
252
+ "validator_feedback_bridge.adapter.lora_embedding_B",
253
+ "validator_feedback_bridge.adapter.lora_magnitude_vector",
254
+ "validator_feedback_loop",
255
+ "validator_feedback_loop.rollback_adapter",
256
+ "validator_feedback_loop.rollback_adapter.base_layer",
257
+ "validator_feedback_loop.rollback_adapter.lora_A",
258
+ "validator_feedback_loop.rollback_adapter.lora_A.default",
259
+ "validator_feedback_loop.rollback_adapter.lora_B",
260
+ "validator_feedback_loop.rollback_adapter.lora_B.default",
261
+ "validator_feedback_loop.rollback_adapter.lora_dropout",
262
+ "validator_feedback_loop.rollback_adapter.lora_dropout.default",
263
+ "validator_feedback_loop.rollback_adapter.lora_embedding_A",
264
+ "validator_feedback_loop.rollback_adapter.lora_embedding_B",
265
+ "validator_feedback_loop.rollback_adapter.lora_magnitude_vector",
266
+ "validator_receipt_bridge",
267
+ "validator_receipt_bridge.adapter",
268
+ "validator_uncertainty_bridge",
269
+ "validator_uncertainty_bridge.adapter",
270
+ "workspace_snapshot_manager",
271
+ "workspace_snapshot_manager.rollback_context",
272
+ "worktree_binding",
273
+ "worktree_binding.adapter",
274
+ "worktree_manager",
275
+ "worktree_manager.adapter"
276
+ ],
277
+ "trainable_parameter_report": {
278
+ "backbone_frozen": true,
279
+ "backbone_present": true,
280
+ "hidden_size": 2048,
281
+ "host_parameters": 176785413,
282
+ "host_trainable_parameters": 524288,
283
+ "total_parameters": 34132843141,
284
+ "trainable_parameters": 524288,
285
+ "version": "veriloop.coder_peft_host.v1"
286
+ },
287
+ "version": "veriloop.coder_peft_host.v1"
288
+ }
rollback_adapter/rollback_adapter_plan.json ADDED
@@ -0,0 +1,251 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "backbone": "/public/wang_libo/veriloop_coder_e1/model",
3
+ "dataset_summary": {
4
+ "cause_vocab": [
5
+ "none",
6
+ "evidence_contradiction",
7
+ "execution_negation",
8
+ "validator_rejection",
9
+ "patch_regression",
10
+ "worktree_conflict",
11
+ "high_risk_unsupported",
12
+ "tool_selfcheck_failure",
13
+ "reverse_engineering_speculation"
14
+ ],
15
+ "eval_size": 45,
16
+ "mode_vocab": [
17
+ "contradiction",
18
+ "execution_failure",
19
+ "validator_rejection",
20
+ "patch_regression",
21
+ "worktree_conflict",
22
+ "high_risk_unsupported",
23
+ "tool_selfcheck_repair",
24
+ "reverse_engineering_drift",
25
+ "validated_stable"
26
+ ],
27
+ "modes": [
28
+ "contradiction",
29
+ "execution_failure",
30
+ "high_risk_unsupported",
31
+ "patch_regression",
32
+ "reverse_engineering_drift",
33
+ "tool_selfcheck_repair",
34
+ "validated_stable",
35
+ "validator_rejection",
36
+ "worktree_conflict"
37
+ ],
38
+ "need_vocab": [
39
+ "none",
40
+ "advisory",
41
+ "required"
42
+ ],
43
+ "rewrite_vocab": [
44
+ "do_not_rollback",
45
+ "local_patch_only",
46
+ "validator_scoped_revision",
47
+ "minimum_diff_patch",
48
+ "do_not_overwrite_worktree_state",
49
+ "fail_closed",
50
+ "repair_via_bounded_selfcheck_loop",
51
+ "bounded_observation_first",
52
+ "local_conflict_preserving_revision"
53
+ ],
54
+ "train_size": 126
55
+ },
56
+ "excluded_surfaces": [
57
+ "(^|\\.)lm_head($|\\.)::Do not retune final token head; too broad and evaluation-heavy.",
58
+ "(^|\\.)embed_tokens($|\\.)::Embedding surgery risks broad semantic drift.",
59
+ "(^|\\.)norm($|\\.)::Global norm tuning can destabilize calibration across scenes.",
60
+ "attnres|attention_residual::Block AttnRes may be mounted structurally but is never a PEFT target.",
61
+ "dualpath::DualPath is serving/runtime infrastructure only.",
62
+ "mhc|hyper[-_]?connection::mHC-inspired stability hooks remain structural, not PEFT surfaces.",
63
+ "rope|rotary::RoPE/context surgery is handled architecturally, not by narrow PEFT here.",
64
+ "kvcache|kv_cache::KV-cache runtime surfaces are not PEFT targets.",
65
+ "(^|\\.)memory(_store|_bank)?($|\\.)::Persistent memory stores are harness/runtime policy surfaces, not PEFT targets."
66
+ ],
67
+ "notes": [
68
+ "Primary route is host-surface-first rollback training.",
69
+ "Validator receipts, failure routing, patch state, worktree continuity, self-check repair, and reverse-engineering boundedness are first-class signals.",
70
+ "DualPath, Block AttnRes, mHC hooks, and MoE routers/experts remain structurally excluded.",
71
+ "This adapter should improve rollback obedience, not general coding free-formity."
72
+ ],
73
+ "peft_method": "lora_narrow",
74
+ "product_line": "veriloop_coder",
75
+ "recipe": {
76
+ "adapter_family": "validator_rollback",
77
+ "backbone": "/public/wang_libo/veriloop_coder_e1/model",
78
+ "backbone_family": "qwen_dense",
79
+ "excluded_patterns": [
80
+ "(?i)\\bdualpath\\b",
81
+ "(?i)\\bmhc\\b",
82
+ "(?i)\\bfull[_\\- ]?attnres\\b",
83
+ "(?i)\\battnres(_full)?\\b",
84
+ "(?i)\\brouter\\b",
85
+ "(?i)\\bexperts?\\b",
86
+ "(?i)\\bmoe\\b.*\\b(gate|router|expert)\\b",
87
+ "(?i)\\brope\\b.*\\b(freq|inv_freq|theta|rotary)\\b",
88
+ "(?i)\\bkvcache\\b",
89
+ "(?i)\\bposition_embedding\\b",
90
+ "(?i)\\bembed(tokens|ding)?\\b",
91
+ "(?i)\\blm_head\\b"
92
+ ],
93
+ "harness_constraints": [
94
+ "Harness Engineering remains the primary convergence layer.",
95
+ "Adapter must not bypass runtime orchestrator / validator / rollback loops.",
96
+ "Adapter outputs remain subordinate to VeriLoop control-plane decisions.",
97
+ "Adapter must not create hidden prompt-style memory authority.",
98
+ "Adapter must preserve validator-first correction and minimal-diff rollback discipline.",
99
+ "Patch production must remain subordinate to validator receipts and failure routing."
100
+ ],
101
+ "hyperparams": {
102
+ "alpha": 16,
103
+ "bias": "none",
104
+ "dropout": 0.08,
105
+ "fan_in_fan_out": false,
106
+ "modules_to_save": [
107
+ "input_layernorm"
108
+ ],
109
+ "r": 8,
110
+ "task_type": "CAUSAL_LM"
111
+ },
112
+ "merge_policy": "side_load",
113
+ "metadata": {
114
+ "allow_backbone_bridge": false,
115
+ "allow_vla_action_expert": false,
116
+ "harness_first": true,
117
+ "prefer_explicit_heads": true,
118
+ "prefer_qlora_for_backbone_bridge": false,
119
+ "require_harness_first": true,
120
+ "reverse_engineering_readiness": true,
121
+ "rollback_training": true,
122
+ "selector_group_count": 2,
123
+ "selfcheck_loop_readiness": true,
124
+ "strict_narrow_scope": true,
125
+ "trainer": "veriloop.rollback_adapter_trainer.v9.qwen36"
126
+ },
127
+ "notes": [
128
+ "Backbone bridge tuning disabled explicitly; selector stays on custom surfaces or no-op.",
129
+ "Backbone family inferred as qwen_dense.",
130
+ "PEFT method resolved as lora_narrow.",
131
+ "Recipe is harness-first: runtime convergence remains in VeriLoop control-plane + harness, not in broad weight surgery.",
132
+ "Block AttnRes, DualPath, mHC hooks, RoPE, KV-cache, and broad MoE routing remain structurally excluded."
133
+ ],
134
+ "peft_method": "lora_narrow",
135
+ "precision_policy": "auto",
136
+ "product_line": "veriloop_coder",
137
+ "regression_requirements": [
138
+ "Must pass PEFT regression guard structural policy checks.",
139
+ "Must not introduce forbidden backbone/serving structural targets.",
140
+ "Must preserve harness regression envelope for the selected product line.",
141
+ "Validator feedback uptake must not regress.",
142
+ "Rollback minimality and revision fidelity must not regress."
143
+ ],
144
+ "target_groups": [
145
+ {
146
+ "alpha": 16,
147
+ "dropout": 0.05,
148
+ "name": "group_1_custom_runtime_harness_bridge",
149
+ "rank": 8,
150
+ "rationale": "Runtime / harness obedience should attach to explicit interface bridges before any backbone fallback.",
151
+ "surface": "custom_runtime_harness_bridge",
152
+ "target_modules": [
153
+ "failure_signal_bridge.rollback_bridge",
154
+ "request_normalizer",
155
+ "request_normalizer.adapter",
156
+ "rollback_adapter",
157
+ "rollback_adapter.head",
158
+ "runtime_harness_adapter",
159
+ "runtime_harness_adapter.bridge",
160
+ "tool_protocol_adapter",
161
+ "tool_protocol_adapter.bridge",
162
+ "toolspec_bridge",
163
+ "toolspec_bridge.adapter",
164
+ "toolspec_head",
165
+ "toolspec_head.param_schema_adapter",
166
+ "toolspec_head.postcondition_adapter",
167
+ "toolspec_head.precondition_adapter",
168
+ "toolspec_head.receipt_formatter",
169
+ "toolspec_head.trigger_gate",
170
+ "toolspec_head.validator_gate",
171
+ "validator_feedback_loop.rollback_adapter"
172
+ ]
173
+ },
174
+ {
175
+ "alpha": 16,
176
+ "dropout": 0.0,
177
+ "name": "group_2_custom_validator_bridge",
178
+ "rank": 8,
179
+ "rationale": "Validation and rollback fidelity should prefer explicit validator / rollback bridges.",
180
+ "surface": "custom_validator_bridge",
181
+ "target_modules": [
182
+ "failure_signal_bridge",
183
+ "failure_signal_bridge.rollback_bridge",
184
+ "rollback_adapter",
185
+ "rollback_adapter.head",
186
+ "sandbox_rollback_bridge",
187
+ "sandbox_rollback_bridge.adapter",
188
+ "validator_feedback_bridge",
189
+ "validator_feedback_bridge.adapter",
190
+ "validator_feedback_loop.rollback_adapter"
191
+ ]
192
+ }
193
+ ],
194
+ "target_modules": [
195
+ "failure_signal_bridge.rollback_bridge",
196
+ "request_normalizer",
197
+ "request_normalizer.adapter",
198
+ "rollback_adapter",
199
+ "rollback_adapter.head",
200
+ "runtime_harness_adapter",
201
+ "runtime_harness_adapter.bridge",
202
+ "tool_protocol_adapter",
203
+ "tool_protocol_adapter.bridge",
204
+ "toolspec_bridge",
205
+ "toolspec_bridge.adapter",
206
+ "toolspec_head",
207
+ "toolspec_head.param_schema_adapter",
208
+ "toolspec_head.postcondition_adapter",
209
+ "toolspec_head.precondition_adapter",
210
+ "toolspec_head.receipt_formatter",
211
+ "toolspec_head.trigger_gate",
212
+ "toolspec_head.validator_gate",
213
+ "validator_feedback_loop.rollback_adapter",
214
+ "failure_signal_bridge",
215
+ "sandbox_rollback_bridge",
216
+ "sandbox_rollback_bridge.adapter",
217
+ "validator_feedback_bridge",
218
+ "validator_feedback_bridge.adapter"
219
+ ],
220
+ "version": "veriloop.lora_recipe_veriloop.v2"
221
+ },
222
+ "selected_surfaces": [
223
+ "custom_runtime_harness_bridge",
224
+ "custom_validator_bridge",
225
+ "custom_memory_boundary_bridge"
226
+ ],
227
+ "selected_target_modules": [
228
+ "failure_signal_bridge.rollback_bridge",
229
+ "rollback_adapter.head",
230
+ "runtime_harness_adapter.bridge",
231
+ "sandbox_rollback_bridge.adapter",
232
+ "toolspec_head.validator_gate",
233
+ "validator_feedback_bridge.adapter",
234
+ "validator_feedback_loop.rollback_adapter",
235
+ "memory_boundary_guard.adapter",
236
+ "memory_boundary_guard.rollback_filter",
237
+ "request_normalizer.adapter",
238
+ "tool_protocol_adapter.bridge",
239
+ "toolspec_bridge.adapter",
240
+ "toolspec_head.postcondition_adapter",
241
+ "toolspec_head.precondition_adapter",
242
+ "toolspec_head.receipt_formatter",
243
+ "toolspec_head.trigger_gate"
244
+ ],
245
+ "selection_mode": "minimal",
246
+ "version": "veriloop.rollback_adapter_trainer.v9.qwen36",
247
+ "warnings": [
248
+ "Harness Engineering is primary; PEFT is limited to obedience-facing, interface-facing support surfaces.",
249
+ "Backbone bridge tuning disabled explicitly; selector stays on custom surfaces or no-op."
250
+ ]
251
+ }
rollback_adapter/rollback_adapter_train_result.json ADDED
@@ -0,0 +1,363 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "artifacts": {
3
+ "adapter_dir": "/private/wang_libo/veriloop_coder_e1/outputs/rollback_qwen36_rootfix_run1/adapter",
4
+ "best_checkpoint_manifest": "/private/wang_libo/veriloop_coder_e1/outputs/rollback_qwen36_rootfix_run1/best_checkpoint_manifest.json",
5
+ "epoch_history": "/private/wang_libo/veriloop_coder_e1/outputs/rollback_qwen36_rootfix_run1/epoch_history.json",
6
+ "eval_jsonl": "/private/wang_libo/veriloop_coder_e1/outputs/rollback_qwen36_rootfix_run1/rollback_eval.jsonl",
7
+ "host_manifest": "/private/wang_libo/veriloop_coder_e1/outputs/rollback_qwen36_rootfix_run1/host_manifest.json",
8
+ "plan_json": "/private/wang_libo/veriloop_coder_e1/outputs/rollback_qwen36_rootfix_run1/rollback_adapter_plan.json",
9
+ "rollback_head": "/private/wang_libo/veriloop_coder_e1/outputs/rollback_qwen36_rootfix_run1/rollback_head.pt",
10
+ "tokenizer_dir": "/private/wang_libo/veriloop_coder_e1/outputs/rollback_qwen36_rootfix_run1/tokenizer",
11
+ "train_jsonl": "/private/wang_libo/veriloop_coder_e1/outputs/rollback_qwen36_rootfix_run1/rollback_train.jsonl",
12
+ "training_manifest": "/private/wang_libo/veriloop_coder_e1/outputs/rollback_qwen36_rootfix_run1/rollback_training_manifest.json"
13
+ },
14
+ "dataset": {
15
+ "cause_vocab": [
16
+ "none",
17
+ "evidence_contradiction",
18
+ "execution_negation",
19
+ "validator_rejection",
20
+ "patch_regression",
21
+ "worktree_conflict",
22
+ "high_risk_unsupported",
23
+ "tool_selfcheck_failure",
24
+ "reverse_engineering_speculation"
25
+ ],
26
+ "eval_size": 45,
27
+ "mode_vocab": [
28
+ "contradiction",
29
+ "execution_failure",
30
+ "validator_rejection",
31
+ "patch_regression",
32
+ "worktree_conflict",
33
+ "high_risk_unsupported",
34
+ "tool_selfcheck_repair",
35
+ "reverse_engineering_drift",
36
+ "validated_stable"
37
+ ],
38
+ "modes": [
39
+ "contradiction",
40
+ "execution_failure",
41
+ "high_risk_unsupported",
42
+ "patch_regression",
43
+ "reverse_engineering_drift",
44
+ "tool_selfcheck_repair",
45
+ "validated_stable",
46
+ "validator_rejection",
47
+ "worktree_conflict"
48
+ ],
49
+ "need_vocab": [
50
+ "none",
51
+ "advisory",
52
+ "required"
53
+ ],
54
+ "rewrite_vocab": [
55
+ "do_not_rollback",
56
+ "local_patch_only",
57
+ "validator_scoped_revision",
58
+ "minimum_diff_patch",
59
+ "do_not_overwrite_worktree_state",
60
+ "fail_closed",
61
+ "repair_via_bounded_selfcheck_loop",
62
+ "bounded_observation_first",
63
+ "local_conflict_preserving_revision"
64
+ ],
65
+ "train_size": 126
66
+ },
67
+ "eval_metrics": {
68
+ "adapter_exported": true,
69
+ "auto_lora_from_ia3": false,
70
+ "avg_binary_accuracy": 0.8194444444444444,
71
+ "best_epoch": 3,
72
+ "best_quality_score": 0.3986773560841878,
73
+ "cause_accuracy": 0.4888888888888889,
74
+ "count": 45,
75
+ "eval_batches": 45,
76
+ "eval_loss": 7.656409973568387,
77
+ "fail_closed_accuracy": 0.7777777777777778,
78
+ "minimal_diff_accuracy": 0.5555555555555556,
79
+ "mode_accuracy": 0.3333333333333333,
80
+ "need_accuracy": 0.7777777777777778,
81
+ "peft_method": "lora_narrow",
82
+ "preserve_conflict_accuracy": 0.7777777777777778,
83
+ "quality_score": 0.3986773560841878,
84
+ "reverse_engineering_bounded_accuracy": 0.8888888888888888,
85
+ "rewrite_accuracy": 0.4888888888888889,
86
+ "selfcheck_route_accuracy": 0.5555555555555556,
87
+ "session_continuity_accuracy": 1.0,
88
+ "used_peft": true,
89
+ "validator_aligned_accuracy": 1.0,
90
+ "worktree_safe_accuracy": 1.0
91
+ },
92
+ "plan": {
93
+ "backbone": "/public/wang_libo/veriloop_coder_e1/model",
94
+ "dataset_summary": {
95
+ "cause_vocab": [
96
+ "none",
97
+ "evidence_contradiction",
98
+ "execution_negation",
99
+ "validator_rejection",
100
+ "patch_regression",
101
+ "worktree_conflict",
102
+ "high_risk_unsupported",
103
+ "tool_selfcheck_failure",
104
+ "reverse_engineering_speculation"
105
+ ],
106
+ "eval_size": 45,
107
+ "mode_vocab": [
108
+ "contradiction",
109
+ "execution_failure",
110
+ "validator_rejection",
111
+ "patch_regression",
112
+ "worktree_conflict",
113
+ "high_risk_unsupported",
114
+ "tool_selfcheck_repair",
115
+ "reverse_engineering_drift",
116
+ "validated_stable"
117
+ ],
118
+ "modes": [
119
+ "contradiction",
120
+ "execution_failure",
121
+ "high_risk_unsupported",
122
+ "patch_regression",
123
+ "reverse_engineering_drift",
124
+ "tool_selfcheck_repair",
125
+ "validated_stable",
126
+ "validator_rejection",
127
+ "worktree_conflict"
128
+ ],
129
+ "need_vocab": [
130
+ "none",
131
+ "advisory",
132
+ "required"
133
+ ],
134
+ "rewrite_vocab": [
135
+ "do_not_rollback",
136
+ "local_patch_only",
137
+ "validator_scoped_revision",
138
+ "minimum_diff_patch",
139
+ "do_not_overwrite_worktree_state",
140
+ "fail_closed",
141
+ "repair_via_bounded_selfcheck_loop",
142
+ "bounded_observation_first",
143
+ "local_conflict_preserving_revision"
144
+ ],
145
+ "train_size": 126
146
+ },
147
+ "excluded_surfaces": [
148
+ "(^|\\.)lm_head($|\\.)::Do not retune final token head; too broad and evaluation-heavy.",
149
+ "(^|\\.)embed_tokens($|\\.)::Embedding surgery risks broad semantic drift.",
150
+ "(^|\\.)norm($|\\.)::Global norm tuning can destabilize calibration across scenes.",
151
+ "attnres|attention_residual::Block AttnRes may be mounted structurally but is never a PEFT target.",
152
+ "dualpath::DualPath is serving/runtime infrastructure only.",
153
+ "mhc|hyper[-_]?connection::mHC-inspired stability hooks remain structural, not PEFT surfaces.",
154
+ "rope|rotary::RoPE/context surgery is handled architecturally, not by narrow PEFT here.",
155
+ "kvcache|kv_cache::KV-cache runtime surfaces are not PEFT targets.",
156
+ "(^|\\.)memory(_store|_bank)?($|\\.)::Persistent memory stores are harness/runtime policy surfaces, not PEFT targets."
157
+ ],
158
+ "notes": [
159
+ "Primary route is host-surface-first rollback training.",
160
+ "Validator receipts, failure routing, patch state, worktree continuity, self-check repair, and reverse-engineering boundedness are first-class signals.",
161
+ "DualPath, Block AttnRes, mHC hooks, and MoE routers/experts remain structurally excluded.",
162
+ "This adapter should improve rollback obedience, not general coding free-formity."
163
+ ],
164
+ "peft_method": "lora_narrow",
165
+ "product_line": "veriloop_coder",
166
+ "recipe": {
167
+ "adapter_family": "validator_rollback",
168
+ "backbone": "/public/wang_libo/veriloop_coder_e1/model",
169
+ "backbone_family": "qwen_dense",
170
+ "excluded_patterns": [
171
+ "(?i)\\bdualpath\\b",
172
+ "(?i)\\bmhc\\b",
173
+ "(?i)\\bfull[_\\- ]?attnres\\b",
174
+ "(?i)\\battnres(_full)?\\b",
175
+ "(?i)\\brouter\\b",
176
+ "(?i)\\bexperts?\\b",
177
+ "(?i)\\bmoe\\b.*\\b(gate|router|expert)\\b",
178
+ "(?i)\\brope\\b.*\\b(freq|inv_freq|theta|rotary)\\b",
179
+ "(?i)\\bkvcache\\b",
180
+ "(?i)\\bposition_embedding\\b",
181
+ "(?i)\\bembed(tokens|ding)?\\b",
182
+ "(?i)\\blm_head\\b"
183
+ ],
184
+ "harness_constraints": [
185
+ "Harness Engineering remains the primary convergence layer.",
186
+ "Adapter must not bypass runtime orchestrator / validator / rollback loops.",
187
+ "Adapter outputs remain subordinate to VeriLoop control-plane decisions.",
188
+ "Adapter must not create hidden prompt-style memory authority.",
189
+ "Adapter must preserve validator-first correction and minimal-diff rollback discipline.",
190
+ "Patch production must remain subordinate to validator receipts and failure routing."
191
+ ],
192
+ "hyperparams": {
193
+ "alpha": 16,
194
+ "bias": "none",
195
+ "dropout": 0.08,
196
+ "fan_in_fan_out": false,
197
+ "modules_to_save": [
198
+ "input_layernorm"
199
+ ],
200
+ "r": 8,
201
+ "task_type": "CAUSAL_LM"
202
+ },
203
+ "merge_policy": "side_load",
204
+ "metadata": {
205
+ "allow_backbone_bridge": false,
206
+ "allow_vla_action_expert": false,
207
+ "harness_first": true,
208
+ "prefer_explicit_heads": true,
209
+ "prefer_qlora_for_backbone_bridge": false,
210
+ "require_harness_first": true,
211
+ "reverse_engineering_readiness": true,
212
+ "rollback_training": true,
213
+ "selector_group_count": 2,
214
+ "selfcheck_loop_readiness": true,
215
+ "strict_narrow_scope": true,
216
+ "trainer": "veriloop.rollback_adapter_trainer.v9.qwen36"
217
+ },
218
+ "notes": [
219
+ "Backbone bridge tuning disabled explicitly; selector stays on custom surfaces or no-op.",
220
+ "Backbone family inferred as qwen_dense.",
221
+ "PEFT method resolved as lora_narrow.",
222
+ "Recipe is harness-first: runtime convergence remains in VeriLoop control-plane + harness, not in broad weight surgery.",
223
+ "Block AttnRes, DualPath, mHC hooks, RoPE, KV-cache, and broad MoE routing remain structurally excluded."
224
+ ],
225
+ "peft_method": "lora_narrow",
226
+ "precision_policy": "auto",
227
+ "product_line": "veriloop_coder",
228
+ "regression_requirements": [
229
+ "Must pass PEFT regression guard structural policy checks.",
230
+ "Must not introduce forbidden backbone/serving structural targets.",
231
+ "Must preserve harness regression envelope for the selected product line.",
232
+ "Validator feedback uptake must not regress.",
233
+ "Rollback minimality and revision fidelity must not regress."
234
+ ],
235
+ "target_groups": [
236
+ {
237
+ "alpha": 16,
238
+ "dropout": 0.05,
239
+ "name": "group_1_custom_runtime_harness_bridge",
240
+ "rank": 8,
241
+ "rationale": "Runtime / harness obedience should attach to explicit interface bridges before any backbone fallback.",
242
+ "surface": "custom_runtime_harness_bridge",
243
+ "target_modules": [
244
+ "failure_signal_bridge.rollback_bridge",
245
+ "request_normalizer",
246
+ "request_normalizer.adapter",
247
+ "rollback_adapter",
248
+ "rollback_adapter.head",
249
+ "runtime_harness_adapter",
250
+ "runtime_harness_adapter.bridge",
251
+ "tool_protocol_adapter",
252
+ "tool_protocol_adapter.bridge",
253
+ "toolspec_bridge",
254
+ "toolspec_bridge.adapter",
255
+ "toolspec_head",
256
+ "toolspec_head.param_schema_adapter",
257
+ "toolspec_head.postcondition_adapter",
258
+ "toolspec_head.precondition_adapter",
259
+ "toolspec_head.receipt_formatter",
260
+ "toolspec_head.trigger_gate",
261
+ "toolspec_head.validator_gate",
262
+ "validator_feedback_loop.rollback_adapter"
263
+ ]
264
+ },
265
+ {
266
+ "alpha": 16,
267
+ "dropout": 0.0,
268
+ "name": "group_2_custom_validator_bridge",
269
+ "rank": 8,
270
+ "rationale": "Validation and rollback fidelity should prefer explicit validator / rollback bridges.",
271
+ "surface": "custom_validator_bridge",
272
+ "target_modules": [
273
+ "failure_signal_bridge",
274
+ "failure_signal_bridge.rollback_bridge",
275
+ "rollback_adapter",
276
+ "rollback_adapter.head",
277
+ "sandbox_rollback_bridge",
278
+ "sandbox_rollback_bridge.adapter",
279
+ "validator_feedback_bridge",
280
+ "validator_feedback_bridge.adapter",
281
+ "validator_feedback_loop.rollback_adapter"
282
+ ]
283
+ }
284
+ ],
285
+ "target_modules": [
286
+ "failure_signal_bridge.rollback_bridge",
287
+ "request_normalizer",
288
+ "request_normalizer.adapter",
289
+ "rollback_adapter",
290
+ "rollback_adapter.head",
291
+ "runtime_harness_adapter",
292
+ "runtime_harness_adapter.bridge",
293
+ "tool_protocol_adapter",
294
+ "tool_protocol_adapter.bridge",
295
+ "toolspec_bridge",
296
+ "toolspec_bridge.adapter",
297
+ "toolspec_head",
298
+ "toolspec_head.param_schema_adapter",
299
+ "toolspec_head.postcondition_adapter",
300
+ "toolspec_head.precondition_adapter",
301
+ "toolspec_head.receipt_formatter",
302
+ "toolspec_head.trigger_gate",
303
+ "toolspec_head.validator_gate",
304
+ "validator_feedback_loop.rollback_adapter",
305
+ "failure_signal_bridge",
306
+ "sandbox_rollback_bridge",
307
+ "sandbox_rollback_bridge.adapter",
308
+ "validator_feedback_bridge",
309
+ "validator_feedback_bridge.adapter"
310
+ ],
311
+ "version": "veriloop.lora_recipe_veriloop.v2"
312
+ },
313
+ "selected_surfaces": [
314
+ "custom_runtime_harness_bridge",
315
+ "custom_validator_bridge",
316
+ "custom_memory_boundary_bridge"
317
+ ],
318
+ "selected_target_modules": [
319
+ "failure_signal_bridge.rollback_bridge",
320
+ "rollback_adapter.head",
321
+ "runtime_harness_adapter.bridge",
322
+ "sandbox_rollback_bridge.adapter",
323
+ "toolspec_head.validator_gate",
324
+ "validator_feedback_bridge.adapter",
325
+ "validator_feedback_loop.rollback_adapter",
326
+ "memory_boundary_guard.adapter",
327
+ "memory_boundary_guard.rollback_filter",
328
+ "request_normalizer.adapter",
329
+ "tool_protocol_adapter.bridge",
330
+ "toolspec_bridge.adapter",
331
+ "toolspec_head.postcondition_adapter",
332
+ "toolspec_head.precondition_adapter",
333
+ "toolspec_head.receipt_formatter",
334
+ "toolspec_head.trigger_gate"
335
+ ],
336
+ "selection_mode": "minimal",
337
+ "version": "veriloop.rollback_adapter_trainer.v9.qwen36",
338
+ "warnings": [
339
+ "Harness Engineering is primary; PEFT is limited to obedience-facing, interface-facing support surfaces.",
340
+ "Backbone bridge tuning disabled explicitly; selector stays on custom surfaces or no-op."
341
+ ]
342
+ },
343
+ "status": "trained",
344
+ "train_metrics": {
345
+ "adapter_exported": true,
346
+ "auto_lora_from_ia3": false,
347
+ "best_epoch": 3,
348
+ "best_quality_score": 0.3986773560841878,
349
+ "epochs_completed": 4,
350
+ "loss": 0.4452953128114579,
351
+ "micro_batches": 126,
352
+ "micro_batches_total": 504,
353
+ "optimizer_steps": 8,
354
+ "optimizer_steps_total": 32,
355
+ "peft_method": "lora_narrow",
356
+ "used_peft": true
357
+ },
358
+ "version": "veriloop.rollback_adapter_trainer.v9.qwen36",
359
+ "warnings": [
360
+ "Harness Engineering is primary; PEFT is limited to obedience-facing, interface-facing support surfaces.",
361
+ "Backbone bridge tuning disabled explicitly; selector stays on custom surfaces or no-op."
362
+ ]
363
+ }
rollback_adapter/rollback_eval.jsonl ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"mode": "contradiction", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=contradiction\nrollback_focus=eval_conflict_preserve\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nCurrent evidence negates an active claim and the contradiction edge must remain visible during rollback.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nevidence_conflict=explicit\nconflict_edge=preserve\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=retrieved_receipt_conflict\ndominant_cause_signal=evidence_contradiction\nrequired_rewrite_scope=local_conflict_preserving_revision\nforbidden_rewrite_scope=silent_conflict_erasure\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "evidence_contradiction", "rewrite_name": "local_conflict_preserving_revision", "need_id": 2, "cause_id": 1, "rewrite_id": 8, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 0}}
2
+ {"mode": "contradiction", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=contradiction\nrollback_focus=eval_conflict_local\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nA validator-visible contradiction has appeared and local conflict-preserving revision is required.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=minimal\nworkspace_delta=preserved\nrollback_budget=guarded\nclaim_status=locally_demote\nconflict_visibility=required\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=conflicting_ground_truth_edges\ndominant_cause_signal=evidence_contradiction\nrequired_rewrite_scope=preserve_conflict_then_revise_locally\nforbidden_rewrite_scope=broad_regeneration\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "evidence_contradiction", "rewrite_name": "local_conflict_preserving_revision", "need_id": 2, "cause_id": 1, "rewrite_id": 8, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 1}}
3
+ {"mode": "contradiction", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=contradiction\nrollback_focus=eval_conflict_preserve\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nCurrent evidence negates an active claim and the contradiction edge must remain visible during rollback.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nevidence_conflict=explicit\nconflict_edge=preserve\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=retrieved_receipt_conflict\ndominant_cause_signal=evidence_contradiction\nrequired_rewrite_scope=local_conflict_preserving_revision\nforbidden_rewrite_scope=silent_conflict_erasure\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "evidence_contradiction", "rewrite_name": "local_conflict_preserving_revision", "need_id": 2, "cause_id": 1, "rewrite_id": 8, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 0}}
4
+ {"mode": "contradiction", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=contradiction\nrollback_focus=eval_conflict_local\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nA validator-visible contradiction has appeared and local conflict-preserving revision is required.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=minimal\nworkspace_delta=preserved\nrollback_budget=guarded\nclaim_status=locally_demote\nconflict_visibility=required\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=conflicting_ground_truth_edges\ndominant_cause_signal=evidence_contradiction\nrequired_rewrite_scope=preserve_conflict_then_revise_locally\nforbidden_rewrite_scope=broad_regeneration\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "evidence_contradiction", "rewrite_name": "local_conflict_preserving_revision", "need_id": 2, "cause_id": 1, "rewrite_id": 8, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 1}}
5
+ {"mode": "contradiction", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=contradiction\nrollback_focus=eval_conflict_preserve\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nCurrent evidence negates an active claim and the contradiction edge must remain visible during rollback.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nevidence_conflict=explicit\nconflict_edge=preserve\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=retrieved_receipt_conflict\ndominant_cause_signal=evidence_contradiction\nrequired_rewrite_scope=local_conflict_preserving_revision\nforbidden_rewrite_scope=silent_conflict_erasure\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "evidence_contradiction", "rewrite_name": "local_conflict_preserving_revision", "need_id": 2, "cause_id": 1, "rewrite_id": 8, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 0}}
6
+ {"mode": "execution_failure", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=execution_failure\nrollback_focus=eval_exec_local\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nA bounded execution trace invalidated one branch and rollback should remain local to the failed path.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nexecution_trace=failed\nrepair_scope=failed_branch_only\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=sandbox_execution_failure\ndominant_cause_signal=execution_negation\nrequired_rewrite_scope=local_patch_only\nforbidden_rewrite_scope=full_repo_rewrite\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "execution_negation", "rewrite_name": "local_patch_only", "need_id": 2, "cause_id": 2, "rewrite_id": 1, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 1, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 0}}
7
+ {"mode": "execution_failure", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=execution_failure\nrollback_focus=eval_exec_patch\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nThe latest patch failed during sandbox execution while neighboring validated work should be retained.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=minimal\nworkspace_delta=preserved\nrollback_budget=guarded\nsandbox_status=nonzero\nrepair_path=local_patch\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=bounded_test_failure\ndominant_cause_signal=execution_negation\nrequired_rewrite_scope=failed_branch_local_repair\nforbidden_rewrite_scope=rewrite_validated_regions\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "execution_negation", "rewrite_name": "local_patch_only", "need_id": 2, "cause_id": 2, "rewrite_id": 1, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 1, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 1}}
8
+ {"mode": "execution_failure", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=execution_failure\nrollback_focus=eval_exec_local\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nA bounded execution trace invalidated one branch and rollback should remain local to the failed path.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nexecution_trace=failed\nrepair_scope=failed_branch_only\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=sandbox_execution_failure\ndominant_cause_signal=execution_negation\nrequired_rewrite_scope=local_patch_only\nforbidden_rewrite_scope=full_repo_rewrite\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "execution_negation", "rewrite_name": "local_patch_only", "need_id": 2, "cause_id": 2, "rewrite_id": 1, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 1, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 0}}
9
+ {"mode": "execution_failure", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=execution_failure\nrollback_focus=eval_exec_patch\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nThe latest patch failed during sandbox execution while neighboring validated work should be retained.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=minimal\nworkspace_delta=preserved\nrollback_budget=guarded\nsandbox_status=nonzero\nrepair_path=local_patch\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=bounded_test_failure\ndominant_cause_signal=execution_negation\nrequired_rewrite_scope=failed_branch_local_repair\nforbidden_rewrite_scope=rewrite_validated_regions\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "execution_negation", "rewrite_name": "local_patch_only", "need_id": 2, "cause_id": 2, "rewrite_id": 1, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 1, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 1}}
10
+ {"mode": "execution_failure", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=execution_failure\nrollback_focus=eval_exec_local\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nA bounded execution trace invalidated one branch and rollback should remain local to the failed path.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nexecution_trace=failed\nrepair_scope=failed_branch_only\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=sandbox_execution_failure\ndominant_cause_signal=execution_negation\nrequired_rewrite_scope=local_patch_only\nforbidden_rewrite_scope=full_repo_rewrite\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "execution_negation", "rewrite_name": "local_patch_only", "need_id": 2, "cause_id": 2, "rewrite_id": 1, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 1, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 0}}
11
+ {"mode": "validator_rejection", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=validator_rejection\nrollback_focus=eval_validator_scope\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nValidator receipts reject the current submission and the correction must stay validator-scoped.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nvalidator_gate=reject\nreceipt_alignment=broken\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=validator_receipt_rejection\ndominant_cause_signal=validator_rejection\nrequired_rewrite_scope=validator_scoped_revision\nforbidden_rewrite_scope=receipt_agnostic_rewrite\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "validator_rejection", "rewrite_name": "validator_scoped_revision", "need_id": 2, "cause_id": 3, "rewrite_id": 2, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 1, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 0}}
12
+ {"mode": "validator_rejection", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=validator_rejection\nrollback_focus=eval_validator_receipt\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nAcceptance criteria failed under validator review even though the patch looks superficially plausible.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=minimal\nworkspace_delta=preserved\nrollback_budget=guarded\nacceptance_criteria=failed\nvalidator_scope=authoritative\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=acceptance_gate_reject\ndominant_cause_signal=validator_rejection\nrequired_rewrite_scope=validator_bound_local_revision\nforbidden_rewrite_scope=bypass_validator_scope\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "validator_rejection", "rewrite_name": "validator_scoped_revision", "need_id": 2, "cause_id": 3, "rewrite_id": 2, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 1, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 1}}
13
+ {"mode": "validator_rejection", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=validator_rejection\nrollback_focus=eval_validator_scope\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nValidator receipts reject the current submission and the correction must stay validator-scoped.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nvalidator_gate=reject\nreceipt_alignment=broken\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=validator_receipt_rejection\ndominant_cause_signal=validator_rejection\nrequired_rewrite_scope=validator_scoped_revision\nforbidden_rewrite_scope=receipt_agnostic_rewrite\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "validator_rejection", "rewrite_name": "validator_scoped_revision", "need_id": 2, "cause_id": 3, "rewrite_id": 2, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 1, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 0}}
14
+ {"mode": "validator_rejection", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=validator_rejection\nrollback_focus=eval_validator_receipt\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nAcceptance criteria failed under validator review even though the patch looks superficially plausible.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=minimal\nworkspace_delta=preserved\nrollback_budget=guarded\nacceptance_criteria=failed\nvalidator_scope=authoritative\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=acceptance_gate_reject\ndominant_cause_signal=validator_rejection\nrequired_rewrite_scope=validator_bound_local_revision\nforbidden_rewrite_scope=bypass_validator_scope\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "validator_rejection", "rewrite_name": "validator_scoped_revision", "need_id": 2, "cause_id": 3, "rewrite_id": 2, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 1, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 1}}
15
+ {"mode": "validator_rejection", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=validator_rejection\nrollback_focus=eval_validator_scope\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nValidator receipts reject the current submission and the correction must stay validator-scoped.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nvalidator_gate=reject\nreceipt_alignment=broken\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=validator_receipt_rejection\ndominant_cause_signal=validator_rejection\nrequired_rewrite_scope=validator_scoped_revision\nforbidden_rewrite_scope=receipt_agnostic_rewrite\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "validator_rejection", "rewrite_name": "validator_scoped_revision", "need_id": 2, "cause_id": 3, "rewrite_id": 2, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 1, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 0}}
16
+ {"mode": "patch_regression", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=patch_regression\nrollback_focus=eval_patch_smallest\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nOne small region regressed after the newest patch and only that region should be revised.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nregression_scope=localized\nminimal_diff=required\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=patch_regression_observed\ndominant_cause_signal=patch_regression\nrequired_rewrite_scope=minimum_diff_patch\nforbidden_rewrite_scope=wide_area_rewrite\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "advisory", "cause_name": "patch_regression", "rewrite_name": "minimum_diff_patch", "need_id": 1, "cause_id": 4, "rewrite_id": 3, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 1, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 0}}
17
+ {"mode": "patch_regression", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=patch_regression\nrollback_focus=eval_patch_local\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nA local regression appeared in the edited diff while previously validated segments remain good.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=minimal\nworkspace_delta=preserved\nrollback_budget=guarded\npatch_lineage=mostly_valid\nrollback_scope=single_region\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=recent_diff_regressed\ndominant_cause_signal=patch_regression\nrequired_rewrite_scope=single_region_patch_repair\nforbidden_rewrite_scope=discard_valid_neighbors\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "advisory", "cause_name": "patch_regression", "rewrite_name": "minimum_diff_patch", "need_id": 1, "cause_id": 4, "rewrite_id": 3, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 1, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 1}}
18
+ {"mode": "patch_regression", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=patch_regression\nrollback_focus=eval_patch_smallest\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nOne small region regressed after the newest patch and only that region should be revised.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nregression_scope=localized\nminimal_diff=required\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=patch_regression_observed\ndominant_cause_signal=patch_regression\nrequired_rewrite_scope=minimum_diff_patch\nforbidden_rewrite_scope=wide_area_rewrite\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "advisory", "cause_name": "patch_regression", "rewrite_name": "minimum_diff_patch", "need_id": 1, "cause_id": 4, "rewrite_id": 3, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 1, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 0}}
19
+ {"mode": "patch_regression", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=patch_regression\nrollback_focus=eval_patch_local\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nA local regression appeared in the edited diff while previously validated segments remain good.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=minimal\nworkspace_delta=preserved\nrollback_budget=guarded\npatch_lineage=mostly_valid\nrollback_scope=single_region\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=recent_diff_regressed\ndominant_cause_signal=patch_regression\nrequired_rewrite_scope=single_region_patch_repair\nforbidden_rewrite_scope=discard_valid_neighbors\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "advisory", "cause_name": "patch_regression", "rewrite_name": "minimum_diff_patch", "need_id": 1, "cause_id": 4, "rewrite_id": 3, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 1, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 1}}
20
+ {"mode": "patch_regression", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=patch_regression\nrollback_focus=eval_patch_smallest\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nOne small region regressed after the newest patch and only that region should be revised.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nregression_scope=localized\nminimal_diff=required\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=patch_regression_observed\ndominant_cause_signal=patch_regression\nrequired_rewrite_scope=minimum_diff_patch\nforbidden_rewrite_scope=wide_area_rewrite\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "advisory", "cause_name": "patch_regression", "rewrite_name": "minimum_diff_patch", "need_id": 1, "cause_id": 4, "rewrite_id": 3, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 1, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 0}}
21
+ {"mode": "worktree_conflict", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=worktree_conflict\nrollback_focus=eval_worktree_snapshot\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nThe rollback candidate conflicts with the live worktree snapshot and should not overwrite workspace state.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nworktree_state=conflicted\noverwrite_policy=forbidden\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=live_worktree_conflict\ndominant_cause_signal=worktree_conflict\nrequired_rewrite_scope=do_not_overwrite_worktree_state\nforbidden_rewrite_scope=overwrite_workspace_state\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "worktree_conflict", "rewrite_name": "do_not_overwrite_worktree_state", "need_id": 2, "cause_id": 5, "rewrite_id": 4, "preserve_conflict": 1, "minimal_diff": 0, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 0}}
22
+ {"mode": "worktree_conflict", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=worktree_conflict\nrollback_focus=eval_worktree_preserve\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nWorkspace lineage drift is present, so rollback must preserve current worktree context.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=minimal\nworkspace_delta=preserved\nrollback_budget=guarded\nworkspace_snapshot=lineage_mismatch\nrollback_scope=preserve_worktree\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=snapshot_lineage_mismatch\ndominant_cause_signal=worktree_conflict\nrequired_rewrite_scope=preserve_worktree_then_reconcile\nforbidden_rewrite_scope=destructive_checkout_style_rewrite\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "worktree_conflict", "rewrite_name": "do_not_overwrite_worktree_state", "need_id": 2, "cause_id": 5, "rewrite_id": 4, "preserve_conflict": 1, "minimal_diff": 0, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 1}}
23
+ {"mode": "worktree_conflict", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=worktree_conflict\nrollback_focus=eval_worktree_snapshot\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nThe rollback candidate conflicts with the live worktree snapshot and should not overwrite workspace state.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nworktree_state=conflicted\noverwrite_policy=forbidden\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=live_worktree_conflict\ndominant_cause_signal=worktree_conflict\nrequired_rewrite_scope=do_not_overwrite_worktree_state\nforbidden_rewrite_scope=overwrite_workspace_state\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "worktree_conflict", "rewrite_name": "do_not_overwrite_worktree_state", "need_id": 2, "cause_id": 5, "rewrite_id": 4, "preserve_conflict": 1, "minimal_diff": 0, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 0}}
24
+ {"mode": "worktree_conflict", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=worktree_conflict\nrollback_focus=eval_worktree_preserve\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nWorkspace lineage drift is present, so rollback must preserve current worktree context.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=minimal\nworkspace_delta=preserved\nrollback_budget=guarded\nworkspace_snapshot=lineage_mismatch\nrollback_scope=preserve_worktree\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=snapshot_lineage_mismatch\ndominant_cause_signal=worktree_conflict\nrequired_rewrite_scope=preserve_worktree_then_reconcile\nforbidden_rewrite_scope=destructive_checkout_style_rewrite\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "worktree_conflict", "rewrite_name": "do_not_overwrite_worktree_state", "need_id": 2, "cause_id": 5, "rewrite_id": 4, "preserve_conflict": 1, "minimal_diff": 0, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 1}}
25
+ {"mode": "worktree_conflict", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=worktree_conflict\nrollback_focus=eval_worktree_snapshot\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nThe rollback candidate conflicts with the live worktree snapshot and should not overwrite workspace state.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nworktree_state=conflicted\noverwrite_policy=forbidden\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=live_worktree_conflict\ndominant_cause_signal=worktree_conflict\nrequired_rewrite_scope=do_not_overwrite_worktree_state\nforbidden_rewrite_scope=overwrite_workspace_state\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "worktree_conflict", "rewrite_name": "do_not_overwrite_worktree_state", "need_id": 2, "cause_id": 5, "rewrite_id": 4, "preserve_conflict": 1, "minimal_diff": 0, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 0}}
26
+ {"mode": "high_risk_unsupported", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=high_risk_unsupported\nrollback_focus=eval_fail_closed\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nA high-risk branch lacks evidence and must fail closed instead of continuing.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nrisk_band=high\nsupport_state=insufficient\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=high_risk_without_support\ndominant_cause_signal=high_risk_unsupported\nrequired_rewrite_scope=fail_closed\nforbidden_rewrite_scope=continue_as_validated\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "high_risk_unsupported", "rewrite_name": "fail_closed", "need_id": 2, "cause_id": 6, "rewrite_id": 5, "preserve_conflict": 0, "minimal_diff": 0, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 0, "fail_closed": 1}, "metadata": {"split": "eval", "variant_rank": 0}}
27
+ {"mode": "high_risk_unsupported", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=high_risk_unsupported\nrollback_focus=eval_high_risk_retract\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nUnsupported high-risk output is present and should be retracted under rollback discipline.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=minimal\nworkspace_delta=preserved\nrollback_budget=guarded\npolicy_mode=fail_closed\nunsupported_claims=retract\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=unsafe_claim_without_receipts\ndominant_cause_signal=high_risk_unsupported\nrequired_rewrite_scope=retract_and_fail_closed\nforbidden_rewrite_scope=speculative_completion\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "high_risk_unsupported", "rewrite_name": "fail_closed", "need_id": 2, "cause_id": 6, "rewrite_id": 5, "preserve_conflict": 0, "minimal_diff": 0, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 0, "fail_closed": 1}, "metadata": {"split": "eval", "variant_rank": 1}}
28
+ {"mode": "high_risk_unsupported", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=high_risk_unsupported\nrollback_focus=eval_fail_closed\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nA high-risk branch lacks evidence and must fail closed instead of continuing.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nrisk_band=high\nsupport_state=insufficient\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=high_risk_without_support\ndominant_cause_signal=high_risk_unsupported\nrequired_rewrite_scope=fail_closed\nforbidden_rewrite_scope=continue_as_validated\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "high_risk_unsupported", "rewrite_name": "fail_closed", "need_id": 2, "cause_id": 6, "rewrite_id": 5, "preserve_conflict": 0, "minimal_diff": 0, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 0, "fail_closed": 1}, "metadata": {"split": "eval", "variant_rank": 0}}
29
+ {"mode": "high_risk_unsupported", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=high_risk_unsupported\nrollback_focus=eval_high_risk_retract\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nUnsupported high-risk output is present and should be retracted under rollback discipline.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=minimal\nworkspace_delta=preserved\nrollback_budget=guarded\npolicy_mode=fail_closed\nunsupported_claims=retract\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=unsafe_claim_without_receipts\ndominant_cause_signal=high_risk_unsupported\nrequired_rewrite_scope=retract_and_fail_closed\nforbidden_rewrite_scope=speculative_completion\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "high_risk_unsupported", "rewrite_name": "fail_closed", "need_id": 2, "cause_id": 6, "rewrite_id": 5, "preserve_conflict": 0, "minimal_diff": 0, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 0, "fail_closed": 1}, "metadata": {"split": "eval", "variant_rank": 1}}
30
+ {"mode": "high_risk_unsupported", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=high_risk_unsupported\nrollback_focus=eval_fail_closed\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nA high-risk branch lacks evidence and must fail closed instead of continuing.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nrisk_band=high\nsupport_state=insufficient\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=high_risk_without_support\ndominant_cause_signal=high_risk_unsupported\nrequired_rewrite_scope=fail_closed\nforbidden_rewrite_scope=continue_as_validated\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "high_risk_unsupported", "rewrite_name": "fail_closed", "need_id": 2, "cause_id": 6, "rewrite_id": 5, "preserve_conflict": 0, "minimal_diff": 0, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 0, "fail_closed": 1}, "metadata": {"split": "eval", "variant_rank": 0}}
31
+ {"mode": "tool_selfcheck_repair", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=tool_selfcheck_repair\nrollback_focus=eval_selfcheck_loop\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nA bounded self-check failed and the system should repair via a narrow self-check loop.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nselfcheck_status=failed\nrepair_loop=bounded\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=tool_selfcheck_negative\ndominant_cause_signal=tool_selfcheck_failure\nrequired_rewrite_scope=repair_via_bounded_selfcheck_loop\nforbidden_rewrite_scope=broad_rewrite_without_selfcheck\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "tool_selfcheck_failure", "rewrite_name": "repair_via_bounded_selfcheck_loop", "need_id": 2, "cause_id": 7, "rewrite_id": 6, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 1, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 0}}
32
+ {"mode": "tool_selfcheck_repair", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=tool_selfcheck_repair\nrollback_focus=eval_selfcheck_patch\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nTool validation rejected the patch and another local self-check repair cycle is warranted.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=minimal\nworkspace_delta=preserved\nrollback_budget=guarded\ntool_validation=negative\nrepair_scope=narrow\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=compiler_or_unit_gate_failed\ndominant_cause_signal=tool_selfcheck_failure\nrequired_rewrite_scope=narrow_selfcheck_repair_cycle\nforbidden_rewrite_scope=skip_repair_loop\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "tool_selfcheck_failure", "rewrite_name": "repair_via_bounded_selfcheck_loop", "need_id": 2, "cause_id": 7, "rewrite_id": 6, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 1, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 1}}
33
+ {"mode": "tool_selfcheck_repair", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=tool_selfcheck_repair\nrollback_focus=eval_selfcheck_loop\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nA bounded self-check failed and the system should repair via a narrow self-check loop.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nselfcheck_status=failed\nrepair_loop=bounded\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=tool_selfcheck_negative\ndominant_cause_signal=tool_selfcheck_failure\nrequired_rewrite_scope=repair_via_bounded_selfcheck_loop\nforbidden_rewrite_scope=broad_rewrite_without_selfcheck\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "tool_selfcheck_failure", "rewrite_name": "repair_via_bounded_selfcheck_loop", "need_id": 2, "cause_id": 7, "rewrite_id": 6, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 1, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 0}}
34
+ {"mode": "tool_selfcheck_repair", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=tool_selfcheck_repair\nrollback_focus=eval_selfcheck_patch\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nTool validation rejected the patch and another local self-check repair cycle is warranted.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=minimal\nworkspace_delta=preserved\nrollback_budget=guarded\ntool_validation=negative\nrepair_scope=narrow\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=compiler_or_unit_gate_failed\ndominant_cause_signal=tool_selfcheck_failure\nrequired_rewrite_scope=narrow_selfcheck_repair_cycle\nforbidden_rewrite_scope=skip_repair_loop\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "tool_selfcheck_failure", "rewrite_name": "repair_via_bounded_selfcheck_loop", "need_id": 2, "cause_id": 7, "rewrite_id": 6, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 1, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 1}}
35
+ {"mode": "tool_selfcheck_repair", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=tool_selfcheck_repair\nrollback_focus=eval_selfcheck_loop\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nA bounded self-check failed and the system should repair via a narrow self-check loop.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nselfcheck_status=failed\nrepair_loop=bounded\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=tool_selfcheck_negative\ndominant_cause_signal=tool_selfcheck_failure\nrequired_rewrite_scope=repair_via_bounded_selfcheck_loop\nforbidden_rewrite_scope=broad_rewrite_without_selfcheck\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "tool_selfcheck_failure", "rewrite_name": "repair_via_bounded_selfcheck_loop", "need_id": 2, "cause_id": 7, "rewrite_id": 6, "preserve_conflict": 1, "minimal_diff": 1, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 1, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 0}}
36
+ {"mode": "reverse_engineering_drift", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=reverse_engineering_drift\nrollback_focus=eval_reverse_observe\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nReverse-engineering inference is drifting beyond observed artifacts and must be pulled back to bounded observation.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nanalysis_mode=reverse_engineering\nobservation_policy=bounded\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=speculative_reverse_engineering\ndominant_cause_signal=reverse_engineering_speculation\nrequired_rewrite_scope=bounded_observation_first\nforbidden_rewrite_scope=invent_hidden_internals\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "reverse_engineering_speculation", "rewrite_name": "bounded_observation_first", "need_id": 2, "cause_id": 8, "rewrite_id": 7, "preserve_conflict": 1, "minimal_diff": 0, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 1, "fail_closed": 1}, "metadata": {"split": "eval", "variant_rank": 0}}
37
+ {"mode": "reverse_engineering_drift", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=reverse_engineering_drift\nrollback_focus=eval_reverse_grounded\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nSpeculative internals appeared during repository analysis and should be rolled back to evidence-grounded structure only.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=minimal\nworkspace_delta=preserved\nrollback_budget=guarded\nspeculation_state=too_high\ngrounding_requirement=observed_only\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=observed_artifacts_insufficient\ndominant_cause_signal=reverse_engineering_speculation\nrequired_rewrite_scope=evidence_grounded_reverse_revision\nforbidden_rewrite_scope=unbounded_architecture_guessing\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "reverse_engineering_speculation", "rewrite_name": "bounded_observation_first", "need_id": 2, "cause_id": 8, "rewrite_id": 7, "preserve_conflict": 1, "minimal_diff": 0, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 1, "fail_closed": 1}, "metadata": {"split": "eval", "variant_rank": 1}}
38
+ {"mode": "reverse_engineering_drift", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=reverse_engineering_drift\nrollback_focus=eval_reverse_observe\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nReverse-engineering inference is drifting beyond observed artifacts and must be pulled back to bounded observation.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nanalysis_mode=reverse_engineering\nobservation_policy=bounded\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=speculative_reverse_engineering\ndominant_cause_signal=reverse_engineering_speculation\nrequired_rewrite_scope=bounded_observation_first\nforbidden_rewrite_scope=invent_hidden_internals\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "reverse_engineering_speculation", "rewrite_name": "bounded_observation_first", "need_id": 2, "cause_id": 8, "rewrite_id": 7, "preserve_conflict": 1, "minimal_diff": 0, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 1, "fail_closed": 1}, "metadata": {"split": "eval", "variant_rank": 0}}
39
+ {"mode": "reverse_engineering_drift", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=reverse_engineering_drift\nrollback_focus=eval_reverse_grounded\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nSpeculative internals appeared during repository analysis and should be rolled back to evidence-grounded structure only.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=minimal\nworkspace_delta=preserved\nrollback_budget=guarded\nspeculation_state=too_high\ngrounding_requirement=observed_only\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=observed_artifacts_insufficient\ndominant_cause_signal=reverse_engineering_speculation\nrequired_rewrite_scope=evidence_grounded_reverse_revision\nforbidden_rewrite_scope=unbounded_architecture_guessing\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "reverse_engineering_speculation", "rewrite_name": "bounded_observation_first", "need_id": 2, "cause_id": 8, "rewrite_id": 7, "preserve_conflict": 1, "minimal_diff": 0, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 1, "fail_closed": 1}, "metadata": {"split": "eval", "variant_rank": 1}}
40
+ {"mode": "reverse_engineering_drift", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=reverse_engineering_drift\nrollback_focus=eval_reverse_observe\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nReverse-engineering inference is drifting beyond observed artifacts and must be pulled back to bounded observation.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nanalysis_mode=reverse_engineering\nobservation_policy=bounded\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=speculative_reverse_engineering\ndominant_cause_signal=reverse_engineering_speculation\nrequired_rewrite_scope=bounded_observation_first\nforbidden_rewrite_scope=invent_hidden_internals\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "required", "cause_name": "reverse_engineering_speculation", "rewrite_name": "bounded_observation_first", "need_id": 2, "cause_id": 8, "rewrite_id": 7, "preserve_conflict": 1, "minimal_diff": 0, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 1, "fail_closed": 1}, "metadata": {"split": "eval", "variant_rank": 0}}
41
+ {"mode": "validated_stable", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=validated_stable\nrollback_focus=eval_no_rollback\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nReceipts and validator outputs agree, so rollback should remain inactive.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nvalidator_gate=pass\nrollback_need=none\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=validator_and_receipts_clean\ndominant_cause_signal=none\nrequired_rewrite_scope=do_not_rollback\nforbidden_rewrite_scope=gratuitous_revision\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "none", "cause_name": "none", "rewrite_name": "do_not_rollback", "need_id": 0, "cause_id": 0, "rewrite_id": 0, "preserve_conflict": 0, "minimal_diff": 0, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 0}}
42
+ {"mode": "validated_stable", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=validated_stable\nrollback_focus=eval_keep_stable\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nThe worktree is stable and validator-clean, so the current revision should be kept.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=minimal\nworkspace_delta=preserved\nrollback_budget=guarded\nreceipt_alignment=clean\nworkspace_state=stable\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=stable_worktree_and_passed_checks\ndominant_cause_signal=none\nrequired_rewrite_scope=keep_current_revision\nforbidden_rewrite_scope=needless_repair_loop\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "none", "cause_name": "none", "rewrite_name": "do_not_rollback", "need_id": 0, "cause_id": 0, "rewrite_id": 0, "preserve_conflict": 0, "minimal_diff": 0, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 1}}
43
+ {"mode": "validated_stable", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=validated_stable\nrollback_focus=eval_no_rollback\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nReceipts and validator outputs agree, so rollback should remain inactive.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nvalidator_gate=pass\nrollback_need=none\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=validator_and_receipts_clean\ndominant_cause_signal=none\nrequired_rewrite_scope=do_not_rollback\nforbidden_rewrite_scope=gratuitous_revision\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "none", "cause_name": "none", "rewrite_name": "do_not_rollback", "need_id": 0, "cause_id": 0, "rewrite_id": 0, "preserve_conflict": 0, "minimal_diff": 0, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 0}}
44
+ {"mode": "validated_stable", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=validated_stable\nrollback_focus=eval_keep_stable\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nThe worktree is stable and validator-clean, so the current revision should be kept.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=minimal\nworkspace_delta=preserved\nrollback_budget=guarded\nreceipt_alignment=clean\nworkspace_state=stable\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=stable_worktree_and_passed_checks\ndominant_cause_signal=none\nrequired_rewrite_scope=keep_current_revision\nforbidden_rewrite_scope=needless_repair_loop\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "none", "cause_name": "none", "rewrite_name": "do_not_rollback", "need_id": 0, "cause_id": 0, "rewrite_id": 0, "preserve_conflict": 0, "minimal_diff": 0, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 1}}
45
+ {"mode": "validated_stable", "product_line": "veriloop_coder", "prompt": "[HARNESS_ROLLBACK_ENVELOPE]\nproduct_line=veriloop_coder\nplanner_enabled=true\nvalidator_enabled=true\nruntime_protocol=true\nsession_continuity=true\nworktree_discipline=true\npatch_alignment=true\nsandbox_validation=true\ntool_selfcheck_loop=true\nreverse_engineering_mode=bounded\nrule=Prefer minimal-diff rollback over broad regeneration.\nrule=Preserve explicit conflicts rather than silently washing them out.\nrule=Validator feedback, failure routes, patch state, workspace deltas, and rollback receipts are first-class rollback inputs.\nrule=If code was executed and failed validator checks, prefer bounded self-check repair before broad rewrite.\nrule=If reverse engineering is requested, preserve observed structure and receipts instead of inventing hidden internals.\nrollback_mode=validated_stable\nrollback_focus=eval_no_rollback\n[/HARNESS_ROLLBACK_ENVELOPE]\n\n[USER_QUERY]\nReceipts and validator outputs agree, so rollback should remain inactive.\n[/USER_QUERY]\n\n[ROLLBACK_CONTEXT]\nvalidator_receipt_state=current\npatch_scope=bounded\nworkspace_delta=tracked\nrollback_budget=tight\nvalidator_gate=pass\nrollback_need=none\n[/ROLLBACK_CONTEXT]\n\n[ROLLBACK_DECISION_FACTORS]\nobserved_failure_channel=validator_and_receipts_clean\ndominant_cause_signal=none\nrequired_rewrite_scope=do_not_rollback\nforbidden_rewrite_scope=gratuitous_revision\n[/ROLLBACK_DECISION_FACTORS]", "labels": {"need_name": "none", "cause_name": "none", "rewrite_name": "do_not_rollback", "need_id": 0, "cause_id": 0, "rewrite_id": 0, "preserve_conflict": 0, "minimal_diff": 0, "validator_aligned": 1, "session_continuity": 1, "worktree_safe": 1, "selfcheck_route": 0, "reverse_engineering_bounded": 0, "fail_closed": 0}, "metadata": {"split": "eval", "variant_rank": 0}}
rollback_adapter/rollback_head.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9a1bc1ea9910f92a75240db4240a85e3aaf92baa3476a65a8cbbd846ab43f4e4
3
+ size 352958879
rollback_adapter/rollback_train.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
rollback_adapter/rollback_training_manifest.json ADDED
@@ -0,0 +1,121 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "adapter_exported": true,
3
+ "dataset_summary": {
4
+ "cause_vocab": [
5
+ "none",
6
+ "evidence_contradiction",
7
+ "execution_negation",
8
+ "validator_rejection",
9
+ "patch_regression",
10
+ "worktree_conflict",
11
+ "high_risk_unsupported",
12
+ "tool_selfcheck_failure",
13
+ "reverse_engineering_speculation"
14
+ ],
15
+ "eval_size": 45,
16
+ "mode_vocab": [
17
+ "contradiction",
18
+ "execution_failure",
19
+ "validator_rejection",
20
+ "patch_regression",
21
+ "worktree_conflict",
22
+ "high_risk_unsupported",
23
+ "tool_selfcheck_repair",
24
+ "reverse_engineering_drift",
25
+ "validated_stable"
26
+ ],
27
+ "modes": [
28
+ "contradiction",
29
+ "execution_failure",
30
+ "high_risk_unsupported",
31
+ "patch_regression",
32
+ "reverse_engineering_drift",
33
+ "tool_selfcheck_repair",
34
+ "validated_stable",
35
+ "validator_rejection",
36
+ "worktree_conflict"
37
+ ],
38
+ "need_vocab": [
39
+ "none",
40
+ "advisory",
41
+ "required"
42
+ ],
43
+ "rewrite_vocab": [
44
+ "do_not_rollback",
45
+ "local_patch_only",
46
+ "validator_scoped_revision",
47
+ "minimum_diff_patch",
48
+ "do_not_overwrite_worktree_state",
49
+ "fail_closed",
50
+ "repair_via_bounded_selfcheck_loop",
51
+ "bounded_observation_first",
52
+ "local_conflict_preserving_revision"
53
+ ],
54
+ "train_size": 126
55
+ },
56
+ "eval_metrics": {
57
+ "adapter_exported": true,
58
+ "auto_lora_from_ia3": false,
59
+ "avg_binary_accuracy": 0.8194444444444444,
60
+ "best_epoch": 3,
61
+ "best_quality_score": 0.3986773560841878,
62
+ "cause_accuracy": 0.4888888888888889,
63
+ "count": 45,
64
+ "eval_batches": 45,
65
+ "eval_loss": 7.656409973568387,
66
+ "fail_closed_accuracy": 0.7777777777777778,
67
+ "minimal_diff_accuracy": 0.5555555555555556,
68
+ "mode_accuracy": 0.3333333333333333,
69
+ "need_accuracy": 0.7777777777777778,
70
+ "peft_method": "lora_narrow",
71
+ "preserve_conflict_accuracy": 0.7777777777777778,
72
+ "quality_score": 0.3986773560841878,
73
+ "reverse_engineering_bounded_accuracy": 0.8888888888888888,
74
+ "rewrite_accuracy": 0.4888888888888889,
75
+ "selfcheck_route_accuracy": 0.5555555555555556,
76
+ "session_continuity_accuracy": 1.0,
77
+ "used_peft": true,
78
+ "validator_aligned_accuracy": 1.0,
79
+ "worktree_safe_accuracy": 1.0
80
+ },
81
+ "load_meta": {
82
+ "chosen_class": "AutoModelForCausalLM",
83
+ "hidden_size": 2048,
84
+ "quantization_mode": "4bit"
85
+ },
86
+ "peft_method": "lora_narrow",
87
+ "selected_target_modules": [
88
+ "failure_signal_bridge.rollback_bridge",
89
+ "rollback_adapter.head",
90
+ "runtime_harness_adapter.bridge",
91
+ "sandbox_rollback_bridge.adapter",
92
+ "toolspec_head.validator_gate",
93
+ "validator_feedback_bridge.adapter",
94
+ "validator_feedback_loop.rollback_adapter",
95
+ "memory_boundary_guard.adapter",
96
+ "memory_boundary_guard.rollback_filter",
97
+ "request_normalizer.adapter",
98
+ "tool_protocol_adapter.bridge",
99
+ "toolspec_bridge.adapter",
100
+ "toolspec_head.postcondition_adapter",
101
+ "toolspec_head.precondition_adapter",
102
+ "toolspec_head.receipt_formatter",
103
+ "toolspec_head.trigger_gate"
104
+ ],
105
+ "status": "trained",
106
+ "train_metrics": {
107
+ "adapter_exported": true,
108
+ "auto_lora_from_ia3": false,
109
+ "best_epoch": 3,
110
+ "best_quality_score": 0.3986773560841878,
111
+ "epochs_completed": 4,
112
+ "loss": 0.4452953128114579,
113
+ "micro_batches": 126,
114
+ "micro_batches_total": 504,
115
+ "optimizer_steps": 8,
116
+ "optimizer_steps_total": 32,
117
+ "peft_method": "lora_narrow",
118
+ "used_peft": true
119
+ },
120
+ "used_peft": true
121
+ }
rollback_adapter/tokenizer/chat_template.jinja ADDED
@@ -0,0 +1,154 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- set image_count = namespace(value=0) %}
2
+ {%- set video_count = namespace(value=0) %}
3
+ {%- macro render_content(content, do_vision_count, is_system_content=false) %}
4
+ {%- if content is string %}
5
+ {{- content }}
6
+ {%- elif content is iterable and content is not mapping %}
7
+ {%- for item in content %}
8
+ {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
9
+ {%- if is_system_content %}
10
+ {{- raise_exception('System message cannot contain images.') }}
11
+ {%- endif %}
12
+ {%- if do_vision_count %}
13
+ {%- set image_count.value = image_count.value + 1 %}
14
+ {%- endif %}
15
+ {%- if add_vision_id %}
16
+ {{- 'Picture ' ~ image_count.value ~ ': ' }}
17
+ {%- endif %}
18
+ {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
19
+ {%- elif 'video' in item or item.type == 'video' %}
20
+ {%- if is_system_content %}
21
+ {{- raise_exception('System message cannot contain videos.') }}
22
+ {%- endif %}
23
+ {%- if do_vision_count %}
24
+ {%- set video_count.value = video_count.value + 1 %}
25
+ {%- endif %}
26
+ {%- if add_vision_id %}
27
+ {{- 'Video ' ~ video_count.value ~ ': ' }}
28
+ {%- endif %}
29
+ {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
30
+ {%- elif 'text' in item %}
31
+ {{- item.text }}
32
+ {%- else %}
33
+ {{- raise_exception('Unexpected item type in content.') }}
34
+ {%- endif %}
35
+ {%- endfor %}
36
+ {%- elif content is none or content is undefined %}
37
+ {{- '' }}
38
+ {%- else %}
39
+ {{- raise_exception('Unexpected content type.') }}
40
+ {%- endif %}
41
+ {%- endmacro %}
42
+ {%- if not messages %}
43
+ {{- raise_exception('No messages provided.') }}
44
+ {%- endif %}
45
+ {%- if tools and tools is iterable and tools is not mapping %}
46
+ {{- '<|im_start|>system\n' }}
47
+ {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
48
+ {%- for tool in tools %}
49
+ {{- "\n" }}
50
+ {{- tool | tojson }}
51
+ {%- endfor %}
52
+ {{- "\n</tools>" }}
53
+ {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
54
+ {%- if messages[0].role == 'system' %}
55
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
56
+ {%- if content %}
57
+ {{- '\n\n' + content }}
58
+ {%- endif %}
59
+ {%- endif %}
60
+ {{- '<|im_end|>\n' }}
61
+ {%- else %}
62
+ {%- if messages[0].role == 'system' %}
63
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
64
+ {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
65
+ {%- endif %}
66
+ {%- endif %}
67
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
68
+ {%- for message in messages[::-1] %}
69
+ {%- set index = (messages|length - 1) - loop.index0 %}
70
+ {%- if ns.multi_step_tool and message.role == "user" %}
71
+ {%- set content = render_content(message.content, false)|trim %}
72
+ {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
73
+ {%- set ns.multi_step_tool = false %}
74
+ {%- set ns.last_query_index = index %}
75
+ {%- endif %}
76
+ {%- endif %}
77
+ {%- endfor %}
78
+ {%- if ns.multi_step_tool %}
79
+ {{- raise_exception('No user query found in messages.') }}
80
+ {%- endif %}
81
+ {%- for message in messages %}
82
+ {%- set content = render_content(message.content, true)|trim %}
83
+ {%- if message.role == "system" %}
84
+ {%- if not loop.first %}
85
+ {{- raise_exception('System message must be at the beginning.') }}
86
+ {%- endif %}
87
+ {%- elif message.role == "user" %}
88
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
89
+ {%- elif message.role == "assistant" %}
90
+ {%- set reasoning_content = '' %}
91
+ {%- if message.reasoning_content is string %}
92
+ {%- set reasoning_content = message.reasoning_content %}
93
+ {%- else %}
94
+ {%- if '</think>' in content %}
95
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
96
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
97
+ {%- endif %}
98
+ {%- endif %}
99
+ {%- set reasoning_content = reasoning_content|trim %}
100
+ {%- if (preserve_thinking is defined and preserve_thinking is true) or (loop.index0 > ns.last_query_index) %}
101
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
102
+ {%- else %}
103
+ {{- '<|im_start|>' + message.role + '\n' + content }}
104
+ {%- endif %}
105
+ {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
106
+ {%- for tool_call in message.tool_calls %}
107
+ {%- if tool_call.function is defined %}
108
+ {%- set tool_call = tool_call.function %}
109
+ {%- endif %}
110
+ {%- if loop.first %}
111
+ {%- if content|trim %}
112
+ {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
113
+ {%- else %}
114
+ {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
115
+ {%- endif %}
116
+ {%- else %}
117
+ {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
118
+ {%- endif %}
119
+ {%- if tool_call.arguments is defined %}
120
+ {%- for args_name, args_value in tool_call.arguments|items %}
121
+ {{- '<parameter=' + args_name + '>\n' }}
122
+ {%- set args_value = args_value | string if args_value is string else args_value | tojson | safe %}
123
+ {{- args_value }}
124
+ {{- '\n</parameter>\n' }}
125
+ {%- endfor %}
126
+ {%- endif %}
127
+ {{- '</function>\n</tool_call>' }}
128
+ {%- endfor %}
129
+ {%- endif %}
130
+ {{- '<|im_end|>\n' }}
131
+ {%- elif message.role == "tool" %}
132
+ {%- if loop.previtem and loop.previtem.role != "tool" %}
133
+ {{- '<|im_start|>user' }}
134
+ {%- endif %}
135
+ {{- '\n<tool_response>\n' }}
136
+ {{- content }}
137
+ {{- '\n</tool_response>' }}
138
+ {%- if not loop.last and loop.nextitem.role != "tool" %}
139
+ {{- '<|im_end|>\n' }}
140
+ {%- elif loop.last %}
141
+ {{- '<|im_end|>\n' }}
142
+ {%- endif %}
143
+ {%- else %}
144
+ {{- raise_exception('Unexpected message role.') }}
145
+ {%- endif %}
146
+ {%- endfor %}
147
+ {%- if add_generation_prompt %}
148
+ {{- '<|im_start|>assistant\n' }}
149
+ {%- if enable_thinking is defined and enable_thinking is false %}
150
+ {{- '<think>\n\n</think>\n\n' }}
151
+ {%- else %}
152
+ {{- '<think>\n' }}
153
+ {%- endif %}
154
+ {%- endif %}
rollback_adapter/tokenizer/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ea2e66b594a0906e9a547c9e6ff9e5fb8a8198439c8cf7d6dc48f23529161223
3
+ size 19989442
rollback_adapter/tokenizer/tokenizer_config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "audio_bos_token": "<|audio_start|>",
4
+ "audio_eos_token": "<|audio_end|>",
5
+ "audio_token": "<|audio_pad|>",
6
+ "backend": "tokenizers",
7
+ "bos_token": null,
8
+ "clean_up_tokenization_spaces": false,
9
+ "eos_token": "<|im_end|>",
10
+ "errors": "replace",
11
+ "image_token": "<|image_pad|>",
12
+ "is_local": true,
13
+ "model_max_length": 262144,
14
+ "model_specific_special_tokens": {
15
+ "audio_bos_token": "<|audio_start|>",
16
+ "audio_eos_token": "<|audio_end|>",
17
+ "audio_token": "<|audio_pad|>",
18
+ "image_token": "<|image_pad|>",
19
+ "video_token": "<|video_pad|>",
20
+ "vision_bos_token": "<|vision_start|>",
21
+ "vision_eos_token": "<|vision_end|>"
22
+ },
23
+ "pad_token": "<|endoftext|>",
24
+ "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
25
+ "split_special_tokens": false,
26
+ "tokenizer_class": "TokenizersBackend",
27
+ "unk_token": null,
28
+ "video_token": "<|video_pad|>",
29
+ "vision_bos_token": "<|vision_start|>",
30
+ "vision_eos_token": "<|vision_end|>"
31
+ }