auto-sync 2026-07-04T05:22:54Z workspace
Browse files- workspace/README.md +26 -13
workspace/README.md
CHANGED
|
@@ -249,14 +249,20 @@ Main executable research pipeline.
|
|
| 249 |
Data/chart export and audits:
|
| 250 |
|
| 251 |
- `scripts/export_cil_charts.py`: exports train/val/test CIL chart DB.
|
| 252 |
-
- `scripts/build_data_accounting.py`: builds data accounting artifacts.
|
| 253 |
-
-
|
|
|
|
|
|
|
| 254 |
- `scripts/audit_leakage.py`: legacy leakage audit.
|
| 255 |
-
- `scripts/audit_action_bounds.py`: action-bound validity audit.
|
|
|
|
| 256 |
- `scripts/audit_chart_feature_sources.py`: audits feature source availability.
|
|
|
|
| 257 |
- `scripts/check_tangent_reconstruction.py`: verifies spline tangent
|
| 258 |
-
reconstruction exactly matches stored residuals.
|
|
|
|
| 259 |
- `scripts/build_action_scale_vector.py`: builds per-dimension action scaling.
|
|
|
|
| 260 |
|
| 261 |
CTT training/proxy/rollout:
|
| 262 |
|
|
@@ -279,13 +285,15 @@ CTT training/proxy/rollout:
|
|
| 279 |
Use `--no-markdown-report` for README-only runs.
|
| 280 |
- `scripts/build_ctt_outcome_gate.py`: measured outcome acceptance gate that
|
| 281 |
combines rollout support and train-clean selector artifacts.
|
| 282 |
-
- `scripts/summarize_ctt_runs.py`: global CSV
|
| 283 |
-
|
| 284 |
|
| 285 |
Dominance and utility:
|
| 286 |
|
| 287 |
-
- `scripts/train_utility_energy.py`: train-only utility energy model.
|
|
|
|
| 288 |
- `scripts/calibrate_dominance.py`: conformal-style dominance calibration rule.
|
|
|
|
| 289 |
- `scripts/eval_dominance_selector.py`: LCB dominance fallback evaluation.
|
| 290 |
Reports selected success, coverage, fallback, unsafe execution, PCCE,
|
| 291 |
selector regret, and support/selector gaps. Use `--no-markdown-report` for
|
|
@@ -686,7 +694,7 @@ Refresh summary:
|
|
| 686 |
.venv/bin/python scripts/summarize_ctt_runs.py \
|
| 687 |
--run-root runs \
|
| 688 |
--out-csv runs/summary_ctt.csv \
|
| 689 |
-
--
|
| 690 |
```
|
| 691 |
|
| 692 |
Run K=16 LCB dominance auto:
|
|
@@ -700,7 +708,8 @@ Run K=16 LCB dominance auto:
|
|
| 700 |
--checkpoint-template 'runs/ctt_residual_base_context_obs_seed{seed}/model.pt' \
|
| 701 |
--out-dir runs/ctt_base_context_obs_dominance_envclip_k16_train_to_test \
|
| 702 |
--k 16 \
|
| 703 |
-
--bootstrap-samples 1000
|
|
|
|
| 704 |
```
|
| 705 |
|
| 706 |
Run leakage audits:
|
|
@@ -709,12 +718,14 @@ Run leakage audits:
|
|
| 709 |
.venv/bin/python scripts/audit_cil_charts.py \
|
| 710 |
--chart-root data/cil_charts \
|
| 711 |
--run-root runs \
|
| 712 |
-
--out-dir runs/leakage_audit
|
|
|
|
| 713 |
|
| 714 |
.venv/bin/python scripts/audit_cil_charts.py \
|
| 715 |
--chart-root data/cil_charts_rgb_refs \
|
| 716 |
--run-root runs \
|
| 717 |
-
--out-dir runs/leakage_audit_rgb_refs
|
|
|
|
| 718 |
```
|
| 719 |
|
| 720 |
Run tangent reconstruction checks:
|
|
@@ -722,11 +733,13 @@ Run tangent reconstruction checks:
|
|
| 722 |
```bash
|
| 723 |
.venv/bin/python scripts/check_tangent_reconstruction.py \
|
| 724 |
--chart-root data/cil_charts \
|
| 725 |
-
--out-dir runs/tangent_reconstruction
|
|
|
|
| 726 |
|
| 727 |
.venv/bin/python scripts/check_tangent_reconstruction.py \
|
| 728 |
--chart-root data/cil_charts_rgb_refs \
|
| 729 |
-
--out-dir runs/tangent_reconstruction_rgb_refs
|
|
|
|
| 730 |
```
|
| 731 |
|
| 732 |
Start HF sync daemon:
|
|
|
|
| 249 |
Data/chart export and audits:
|
| 250 |
|
| 251 |
- `scripts/export_cil_charts.py`: exports train/val/test CIL chart DB.
|
| 252 |
+
- `scripts/build_data_accounting.py`: builds data accounting artifacts. Use
|
| 253 |
+
`--no-markdown-report` for README-only runs.
|
| 254 |
+
- `scripts/audit_cil_charts.py`: leakage audit for chart indexes and run
|
| 255 |
+
hashes. Use `--no-markdown-report` for README-only runs.
|
| 256 |
- `scripts/audit_leakage.py`: legacy leakage audit.
|
| 257 |
+
- `scripts/audit_action_bounds.py`: action-bound validity audit. Use
|
| 258 |
+
`--no-markdown-report` for README-only runs.
|
| 259 |
- `scripts/audit_chart_feature_sources.py`: audits feature source availability.
|
| 260 |
+
Use `--no-markdown-report` for README-only runs.
|
| 261 |
- `scripts/check_tangent_reconstruction.py`: verifies spline tangent
|
| 262 |
+
reconstruction exactly matches stored residuals. Use `--no-markdown-report`
|
| 263 |
+
for README-only runs.
|
| 264 |
- `scripts/build_action_scale_vector.py`: builds per-dimension action scaling.
|
| 265 |
+
Use `--no-markdown-report` for README-only runs.
|
| 266 |
|
| 267 |
CTT training/proxy/rollout:
|
| 268 |
|
|
|
|
| 285 |
Use `--no-markdown-report` for README-only runs.
|
| 286 |
- `scripts/build_ctt_outcome_gate.py`: measured outcome acceptance gate that
|
| 287 |
combines rollout support and train-clean selector artifacts.
|
| 288 |
+
- `scripts/summarize_ctt_runs.py`: global CSV summary. Use
|
| 289 |
+
`--no-markdown-report` so the persistent prose overview remains this README.
|
| 290 |
|
| 291 |
Dominance and utility:
|
| 292 |
|
| 293 |
+
- `scripts/train_utility_energy.py`: train-only utility energy model. Use
|
| 294 |
+
`--no-markdown-report` for README-only runs.
|
| 295 |
- `scripts/calibrate_dominance.py`: conformal-style dominance calibration rule.
|
| 296 |
+
Use `--no-markdown-report` for README-only runs.
|
| 297 |
- `scripts/eval_dominance_selector.py`: LCB dominance fallback evaluation.
|
| 298 |
Reports selected success, coverage, fallback, unsafe execution, PCCE,
|
| 299 |
selector regret, and support/selector gaps. Use `--no-markdown-report` for
|
|
|
|
| 694 |
.venv/bin/python scripts/summarize_ctt_runs.py \
|
| 695 |
--run-root runs \
|
| 696 |
--out-csv runs/summary_ctt.csv \
|
| 697 |
+
--no-markdown-report
|
| 698 |
```
|
| 699 |
|
| 700 |
Run K=16 LCB dominance auto:
|
|
|
|
| 708 |
--checkpoint-template 'runs/ctt_residual_base_context_obs_seed{seed}/model.pt' \
|
| 709 |
--out-dir runs/ctt_base_context_obs_dominance_envclip_k16_train_to_test \
|
| 710 |
--k 16 \
|
| 711 |
+
--bootstrap-samples 1000 \
|
| 712 |
+
--no-markdown-report
|
| 713 |
```
|
| 714 |
|
| 715 |
Run leakage audits:
|
|
|
|
| 718 |
.venv/bin/python scripts/audit_cil_charts.py \
|
| 719 |
--chart-root data/cil_charts \
|
| 720 |
--run-root runs \
|
| 721 |
+
--out-dir runs/leakage_audit \
|
| 722 |
+
--no-markdown-report
|
| 723 |
|
| 724 |
.venv/bin/python scripts/audit_cil_charts.py \
|
| 725 |
--chart-root data/cil_charts_rgb_refs \
|
| 726 |
--run-root runs \
|
| 727 |
+
--out-dir runs/leakage_audit_rgb_refs \
|
| 728 |
+
--no-markdown-report
|
| 729 |
```
|
| 730 |
|
| 731 |
Run tangent reconstruction checks:
|
|
|
|
| 733 |
```bash
|
| 734 |
.venv/bin/python scripts/check_tangent_reconstruction.py \
|
| 735 |
--chart-root data/cil_charts \
|
| 736 |
+
--out-dir runs/tangent_reconstruction \
|
| 737 |
+
--no-markdown-report
|
| 738 |
|
| 739 |
.venv/bin/python scripts/check_tangent_reconstruction.py \
|
| 740 |
--chart-root data/cil_charts_rgb_refs \
|
| 741 |
+
--out-dir runs/tangent_reconstruction_rgb_refs \
|
| 742 |
+
--no-markdown-report
|
| 743 |
```
|
| 744 |
|
| 745 |
Start HF sync daemon:
|