Andrew Park 835b69c59c fix(rescore_swe): never overwrite any row with a no_report result ...

Tighter version of the prior fix: previously we only protected
success=True rows from being downgraded to no_report, but fail-with-real-report
rows could still get clobbered. Result on the advisors qwen36 cell: pass count
preserved, but no_report grew from 56 → 70 (losing real test-failure info
for 14 rows, no net acc improvement either way).

Now: if the new harness result is reason=no_report, keep the old score
regardless of what the old was. Modal harness flakiness is monotone-safe.

2026-05-26 01:57:23 -07:00

..

make_report.py

feat: wire native web_search into GAIA cells across paradigms

2026-05-17 17:55:18 -07:00

rescore_swe.py

fix(rescore_swe): never overwrite any row with a no_report result

2026-05-26 01:57:23 -07:00

run_sweep.py

fix: SWE cloud-loop silent termination + ablation sweep bench label

2026-05-17 18:36:57 -07:00

test_run_sweep.py

chore: ship pending companion files for web_search + worker_pool + ablation

2026-05-18 01:32:28 -07:00