Marius CiepluchandClaude Fable 5.1 47d0f0fac1 bench(unseen): per-commit history cache for ordinary_history; assemble.freeze --history-cache ...

One append-only JSONL per clone (named by the sha256 of its top-level path), one line per commit
with the row as the walk built it and HISTORY_CACHE_VERSION, flushed after every commit so a
killed run keeps its progress. A cached row is reused only when its parents match the current log;
other versions and torn lines are recomputed. With no cache directory nothing changes. The
7,400-commit trigger.dev walk that took over an hour per freeze becomes a one-time cost.
Implemented by Codex (gpt-6-astra), gated here.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

2026-10-13 19:40:34 +02:00

..

pilot

feat(benchmark): freeze the valid-benchmark pilot, 12 verified positives + 240 negatives

2026-10-12 19:42:45 +02:00

train

data: maintainer spot-check, 51 access control locations accepted (7 changes)

2026-10-13 18:09:10 +02:00

__init__.py

feat(benchmark): valid detection benchmark spec + offline modules (groups 1-4)

2026-10-12 18:55:42 +02:00

assemble.py

bench(unseen): per-commit history cache for ordinary_history; assemble.freeze --history-cache

2026-10-13 19:40:34 +02:00

catalog.py

data: hunt training-set candidates (task 4.1), not frozen: 99 verified locations in 28 changes, awaiting spot-check

2026-10-13 16:48:41 +02:00

hunter_arm.py

feat(benchmark): combine --keep-verdicts drops hunter suspicions the classifier rejected

2026-10-13 08:21:59 +02:00

metrics.py

feat(benchmark): valid detection benchmark spec + offline modules (groups 1-4)

2026-10-12 18:55:42 +02:00

semgrep_arm.py

measure: Semgrep arm on the valid pilot, our rules and facts converted

2026-10-13 13:10:19 +02:00

verify.py

feat(benchmark): valid detection benchmark spec + offline modules (groups 1-4)

2026-10-12 18:55:42 +02:00