fix: principled abstention threshold (closes #33) — validated on three question sets - #35
Merged
Merged
Conversation
…three question sets Closes #33. No-hit decision now derives from corpus-relative score statistics + salient-term coverage instead of constants implicitly calibrated to the tuned set's absent questions. Validation: tuned set 2/2 abstain (kept), held-out 2/2 (was 0/2), pre-registered third set 5/6 (bar: >=5/6; pre-registration timestamped before implementation). Covered retrieval preserved: held-out 9/10 top-1, 10/10 top-3. 369 tests; bundles rebuilt. Implemented by codex from spec; independently re-verified here.
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The held-out eval (#34) showed abstention was overfit (0/2 on fresh absent topics). This replaces it with a corpus-relative threshold (score statistics + salient-term coverage, every constant justified in comments), validated with an anti-overfit protocol:
The 'abstains instead of hallucinating' property now holds on questions the threshold has never seen. Implemented by codex (gpt-5.5); independently re-run and verified here.