Chaim
532bef04a7
The pending review queue split items into 'להכרעתך' (chair judgment) vs 'דורש תיקון-חילוץ' (needs re-extraction) by a too-blunt predicate: ANY quality flag + no panel round → re-extraction. That misfiled fact-specific 'application' items (extraction is fine; the open question is whether the rule generalizes) as extraction defects, contradicting the backend, which treats only quote_unverified/truncated_quote/thin_restatement/near_duplicate as defects and routes 'application' to chair judgment. Now isExtractionFixItem checks the backend's DEFECT set, so 'application' (and any other non-defect flag) lands in 'להכרעתך'. For 8508-03-24 this moves 21 application items judgment→correct and leaves 6 thin_restatement as the only re-extraction items. Logic-only fix (predicate), no visual change — design-gate exempt per web-ui/AGENTS.md. tsc + eslint clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>