The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.
RR-GID-CN paper results
Audited jsonl, configs, calibration reports, and generators for 15628925702/RR_GID_CN.
This dataset is the number source for E1–E5. It does not include QMC spill caches (basis_*.dat).
The former standalone repo Guaogua/RR-GID-CN-opt is archived here under opt/ (optimization-track grids; not the freeze snapshot). Paper body numbers still come from freeze_20260920/.
Current paper body: freeze-20260920-ftildev6mix
Tag: freeze-20260920-ftildev6mix. Folder: freeze_20260920/.
This freeze stacks both advisor batches:
- Score from (\tilde\beta) on synthetic E1 and Gas R1; learned VAEAC generator on Gas; (R_{\mathrm{risk}}) near 1 via J=4 and 20 paired replications.
- GAMD design Fisher is (F(\widetilde\beta)+I_S(\widetilde\beta)), not (F(\beta^\star)). E1/E2/E3 were rerun under that protocol.
| path | what |
|---|---|
freeze_20260920/e1_j4_20_ftilde/rows.jsonl |
E1, J=4, 20-rep, design Fisher (F(\widetilde\beta)), SHA256 796719b2ca618dd75a4214f184779f2c84c316faceef32479e8a0d61c4783941 |
freeze_20260920/e2_ftilde_15/rows.jsonl |
E2, 15-rep, four (\alpha), SHA256 8ebdf2165ec7d7357e9fe660bc93a89a080018d0fad68c5857f264238ba750dc |
freeze_20260920/e3_ftilde_t50/rows.jsonl |
E3, t=50 reuse, SHA256 db3b52c587a5cfd1c274c214e1a246cb80cab93bb85984403b6607392d0d5ee1 |
freeze_20260920/risk_gates/fixed_cached_equivalence_order14.json |
E4 order-14 certificate |
freeze_20260920/e5_r1_v6mix_20/rows.jsonl |
Gas R1, scoring from pilot (\tilde\beta), mix (I_S), SHA256 77508171a3cbd2830228e728f39be24341066c3172cc29b34b5025dd5254157b |
freeze_20260920/e5_r2_v6mix_calahat_20/rows.jsonl |
Gas R2, learned generator + calibrated (\widehat A), SHA256 4f30829d3e709b99f26ace6aabfcc894acb630041972347075d8780e96c642b6 |
freeze_20260920/checkpoints/vaeac_gas_opt_v6mix.pt |
Gas VAEAC, SHA256 5bb8d4818bfbf7e6792e9494e2de9bc06eeafa84f795a86221fad03d622583da |
Method note that stays in the paper: E5-R2 canonical rows still use scoring_init: zero. A 20-rep trial that copied R1’s tilde protocol (e5_r2_v6mix_follow_r1_20) lost ranking, so R2 keeps zero init plus calibrated (\widehat A).
See freeze_20260920/FREEZE.json and freeze_20260920/README.md.
Historical freeze (immutable, not current body): freeze-20260914-v6rdv5
Tag: freeze-20260914-v6rdv5. Do not overwrite these jsonl. On this repo they live under paper_results/.
| path | what |
|---|---|
paper_results/e1_synthetic_main/rows.jsonl |
Old E1, J=2, 200 paired replications |
paper_results/e2_nonlinearity/rows.jsonl |
Old E2 |
paper_results/e3_reuse/rows.jsonl |
Old E3 |
paper_results/e5_r1_v6rd/rows.jsonl |
Old Gas R1 (score from 0, kNN generator) |
paper_results/e5_r2_v5/rows.jsonl |
Old Gas R2 |
paper_results/checkpoints/empirical_knn_gas_v1.json |
Frozen empirical kNN handle |
Those files remain on main and on tag freeze-20260914-v6rdv5. Newer work is added under freeze_20260920/ instead of replacing them.
Older still (do not use for paper body)
| path | what |
|---|---|
paper_results/e5_r1_semisynthetic/rows.jsonl |
2026-09-13 R1 freeze |
paper_results/e5_r2_natural/rows.jsonl |
2026-09-13 R2 freeze |
Primary synthetic metric is primary_loss = risk_ratio_raw. Gas metric is family-projection loss in primary_loss.
- Downloads last month
- 203