train-clean bounded tanh CTT 2026-07-03 final report ledger
Browse files
workspace/reports/2026-07-03.md
CHANGED
|
@@ -585,6 +585,10 @@ Action-bound audit status: new rollout artifacts now log action-bound safety lab
|
|
| 585 |
- paper/report/code verification matched local SHA-256 for `workspace/scripts/eval_ctt_generated_rollout.py`, `workspace/scripts/slurm/eval_ctt_generated_rollout.sbatch`, `workspace/tests/test_ctt.py`, `workspace/docs/cil_format.md`, `workspace/latex/main.{tex,pdf}`, `workspace/reports/2026-07-03.md`, and `workspace/reports/draft_audit_ctt.md`.
|
| 586 |
- bounded-tanh run verification matched local SHA-256 for `workspace/runs/ctt_base_context_obs_val_tanh_rollout_comparison/metrics.json`, `workspace/runs/ctt_base_context_obs_test_tanh_rollout_comparison/metrics.json`, `workspace/runs/ctt_base_context_obs_learned_dominance_context_tangent_success_tanh_val_to_test/metrics.json`, and `workspace/runs/ctt_residual_base_context_obs_rollout_test_tanh_seed0/measured_candidates.json`.
|
| 587 |
- representative tanh commits: val/test aggregate `fc50a4d47d1d78dea2596255b866951c73acf44c`, dominance auto `83f5d35d6bbbd5581673688d8e46d16b5d681596`, learned context-tangent-success dominance `4d3e0d2284bd205140377d3c167fdd0025df2a8d`, smoke `08d72b588e1a43e28785568d55ce52553a176ba8`, validation seed dirs `6b5e3dfc51576c6e18bd2dd3f42174eaf39871a0`, `c6aa700587506e8a9db4ea018254698f4576eec1`, `e4f63eba3d70c538c0fd94332fd820fb6ee8dc9a`, and test seed dirs `ab1363b1c6c07a5c90156d3328d4e184ad9f76f0`, `0aba4f599283f1462a4a26126442253a0627d071`, `d80c249cbd193bae01374675d25b3993fa391ca1`.
|
|
|
|
|
|
|
|
|
|
|
|
|
| 588 |
|
| 589 |
## Next Actions
|
| 590 |
|
|
|
|
| 585 |
- paper/report/code verification matched local SHA-256 for `workspace/scripts/eval_ctt_generated_rollout.py`, `workspace/scripts/slurm/eval_ctt_generated_rollout.sbatch`, `workspace/tests/test_ctt.py`, `workspace/docs/cil_format.md`, `workspace/latex/main.{tex,pdf}`, `workspace/reports/2026-07-03.md`, and `workspace/reports/draft_audit_ctt.md`.
|
| 586 |
- bounded-tanh run verification matched local SHA-256 for `workspace/runs/ctt_base_context_obs_val_tanh_rollout_comparison/metrics.json`, `workspace/runs/ctt_base_context_obs_test_tanh_rollout_comparison/metrics.json`, `workspace/runs/ctt_base_context_obs_learned_dominance_context_tangent_success_tanh_val_to_test/metrics.json`, and `workspace/runs/ctt_residual_base_context_obs_rollout_test_tanh_seed0/measured_candidates.json`.
|
| 587 |
- representative tanh commits: val/test aggregate `fc50a4d47d1d78dea2596255b866951c73acf44c`, dominance auto `83f5d35d6bbbd5581673688d8e46d16b5d681596`, learned context-tangent-success dominance `4d3e0d2284bd205140377d3c167fdd0025df2a8d`, smoke `08d72b588e1a43e28785568d55ce52553a176ba8`, validation seed dirs `6b5e3dfc51576c6e18bd2dd3f42174eaf39871a0`, `c6aa700587506e8a9db4ea018254698f4576eec1`, `e4f63eba3d70c538c0fd94332fd820fb6ee8dc9a`, and test seed dirs `ab1363b1c6c07a5c90156d3328d4e184ad9f76f0`, `0aba4f599283f1462a4a26126442253a0627d071`, `d80c249cbd193bae01374675d25b3993fa391ca1`.
|
| 588 |
+
- Train-clean bounded-tanh calibration/dominance targeted HF upload completed and was verified by forced download SHA-256 comparison.
|
| 589 |
+
- uploaded/verified `workspace/latex/main.tex`, `workspace/latex/main.pdf`, `workspace/reports/2026-07-03.md`, `workspace/reports/draft_audit_ctt.md`, `workspace/scripts/slurm/eval_tanh_train_dominance.sbatch`, train-calibration aggregate metrics/candidates, best context-success train-to-test metrics, and seed0 train-calibration measured candidates.
|
| 590 |
+
- representative commits: `latex/main.tex` `29a6a8738a8def04dba89459f7bc2db67a2adeee`, `latex/main.pdf` `3aefc3c18f1d1aabb999c85145a3e6616d10cbe0`, audit report `ae6ed34aa9b1f823f9786bc5e9d080d17f5491fd`, auto dominance `63f89e57e8b2b6a97d750c36737e2af9f6b4a3f7`, learned context-success `2c8670518864590955b9355e4e121e41054dc294`.
|
| 591 |
+
- verified SHA-256 prefixes: `latex/main.tex` `34ca0ce691e9`, `latex/main.pdf` `9b13c1582db8`, train aggregate metrics `80b3d34d0fda`, train aggregate candidates `6842324d24b4`, best learned dominance metrics `03b83b1aaf0e`, seed0 measured candidates `55c972acaab8`.
|
| 592 |
|
| 593 |
## Next Actions
|
| 594 |
|