data: export review-eval ground truth (unblocks the model A/B — qaEngineer#20)#21
Merged
Merged
Conversation
…qaEngineer#20) The 22+ hand-grounded findings were gitignored at data/review-eval/, local to this box — the lab side had only a 3-row truth.seed.jsonl and couldn't score the model A/B without the rest. Committing the real file so the lab can pull it. - `truth.seed.jsonl` — 24 labeled rows with heads resolved to full 40-char SHAs (the short SHAs would not join cleanly against a replay run's full-SHA output). - `README.md` — the schema + the scoring notes that matter: precision must exclude `assertion_only`/`not_grounded` rows, recall must include the `false_negative` rows (the defects the panel missed — #2143, #2210), honesty = disregarded-evidence + fabricated-quote rate. - `build_ds.py` + the original short-SHA file, for provenance. Secret-scanned: labels + claims about our own PRs, nothing sensitive. The gitignore comment said "local until it's big enough to move" — the A/B needs it now, so it moves. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Blocker 2 of the qaEngineer critical path. The 24 hand-grounded findings were gitignored at
data/review-eval/, local to the qaEngineer box — the lab side had only a 3-rowtruth.seed.jsonland can't score the model A/B without the rest. This commits the real file so the lab cangit pullit.What's in it
truth.seed.jsonl— 24 labeled rows, heads resolved to full 40-char SHAs so a replay run's findings join cleanly on(repo, head, file, line). The original short SHAs (7–8 char, inconsistent) would have been a join hazard.README.md— schema + the scoring notes that actually matter:assertion_only/not_groundedrows (findings I didn't independently ground)false_negativerows — the defects the panel missed (protoAgent#2143's orphan, #2210's event-loop major). A replay of the same head should re-find these; a model that doesn't scores a miss.disregarded_evidencecount + fabricated-quote survival ratebuild_ds.py+ the original file, for provenance.Seed: 18 true · 3 false · 1 false-negative · 2 unverified
Secret-scanned — labels and claims about our own PRs, nothing sensitive. The gitignore said "local until it's big enough to move"; the A/B needs it now.
Replay mode (blocker 1, the plugin-side runner) is the next PR.
🤖 Generated with Claude Code