auto-sync 2026-07-03T22:52:00Z workspace (part 5)

workspace/reports/2026-07-03.md

@@ -142,6 +142,10 @@
   - dependent held-out test array `15131733_[0-2]`: 48 test charts, seeds 0/1/2, K=8, same convention.
   - dependent train-calibration array `15131734_[0-2]`: 144 train target charts, same-chart/state source exclusion, seeds 0/1/2, K=8, same convention.
   - dependent aggregate/dominance job `15131740`: after the train/validation/test arrays, aggregates `env_clip` rows and evaluates train-clean dominance variants on held-out test. The dependency chain prevents full rollout if the smoke fails.
+  - first pipeline was canceled while smoke `15131711` was already at ManiSkill env init on node `rc32627`; dependent jobs `15131732`--`15131734` and `15131740` were canceled before running.
+  - retry smoke `15132173` was submitted with `--exclude=rc32627`, same settings, out dir `runs/ctt_residual_base_context_obs_rollout_safety_envclip_smoke_retry1_seed0`.
+  - retry dependent arrays `15132225_[0-2]`, `15132226_[0-2]`, and `15132227_[0-2]` cover validation/test/train-calibration respectively with the same `env_clip` convention.
+  - retry aggregate/dominance job `15132234` depends on the three retry arrays and writes `runs/ctt_base_context_obs_{train_cal,val,test}_envclip_rollout_comparison` plus train-to-test dominance artifacts.
 
 ## Jobs Completed
 
@@ -464,6 +468,8 @@
 - Fix: updated `_scale_actions_4d` to pad shorter scale vectors with `1.0` after action adaptation, added regression test `test_execution_action_scale_vector_pads_after_adapted_dims`, and reran `tests/test_ctt.py` successfully before submitting clean retry arrays `15130732`--`15130734`.
 - Per-dimension train-max retry `15130732`--`15130734` was canceled after two validation seeds completed with an invalid one-value scale vector. Root cause: `sbatch --export` treats commas as separators, so `EXECUTION_ACTION_SCALE_VECTOR=0.3814,...` reached Python as only `0.3814`.
 - Fix: added `EXECUTION_ACTION_SCALE_VECTOR_FILE` to `scripts/slurm/eval_ctt_generated_rollout.sbatch`, removed invalid run dirs, and resubmitted as `15130968`--`15130970`.
+- First `env_clip` smoke `15131711` was canceled manually after appearing to stall at import; its run log later showed it had reached ManiSkill env init on `rc32627`. No measured rows were produced, and dependent jobs `15131732`--`15131734`/`15131740` were canceled before execution.
+- Fix: resubmitted the smoke and dependent rollout/dominance pipeline as `15132173`, `15132225`--`15132227`, and `15132234` with `--exclude=rc32627`.
 
 ## Best Metric Table