fix(extraction): relax records-sentinel gate to a +30 score bonus#88
Merged
Conversation
The hard filter (any records → only sentinel-bearing candidates selectable) caused beautifulsoup to win by forfeit on 231/1497 WCXB pages — 52.8% of the measured selector oracle gap (notes/ selector-oracle-gap-diagnostic.md). Replaced with a single max() over all candidates where sentinel-bearing candidates get SENTINEL_BONUS when records are present; stored score stays raw. Bonus calibration was measured, not guessed: - Offline sweep over 1,497 pages: B=10 (+0.0146) and B=15 (+0.0056) pass per-type gates offline, but BOTH fail the aladin_bestsellers parity hard gate end-to-end — the measured non-sentinel-vs-sentinel gap on that page is 15.84 (rs-trafilatura 145.78 vs beautifulsoup 129.95, query-dependent). - B=30 keeps a 14-point margin against live drift at nearly the same gain as B=20 (+0.0037 vs +0.0046) and zero page-type regressions. Gates at B=30: parity 15/15 (aladin/wanted/hada record cases intact), coding shard 24/24, WCXB 0.814 -> 0.818 (+0.004, matches offline prediction). 98 unit tests (extraction/records/chunking) pass.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Fix A of the selector oracle-gap diagnostic: the hard records-sentinel filter (any repeating-sibling group → only sentinel-bearing candidates selectable) made beautifulsoup win by forfeit on 231/1,497 WCXB pages — 52.8% of the measured 0.068 oracle gap. Replaced with a single
max()over all candidates where sentinel-bearing candidates receiveSENTINEL_BONUS = 30.0when records are present (stored score stays raw; sentinel-driven record-atomic chunking still wins all realistic ties).Bonus calibration — measured, not guessed
Gates (B=30)
Diagnostic + sweep:
notes/selector-oracle-gap-diagnostic.md(local). Together with #86, the selector oracle gap is now 0.809 → 0.818 of a 0.877 ceiling.🤖 Generated with Claude Code