From d0646c1627e79c72a740f23a26ed392fa6609016 Mon Sep 17 00:00:00 2001 From: garrytan Date: Mon, 5 Oct 2026 20:28:40 +0000 Subject: [PATCH] test(office-hours): preservation judge accepts COMPLETED with recorded minor findings --- test/helpers/office-hours-completion.ts | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/test/helpers/office-hours-completion.ts b/test/helpers/office-hours-completion.ts index 9a582f80f..fbe5a6089 100644 --- a/test/helpers/office-hours-completion.ts +++ b/test/helpers/office-hours-completion.ts @@ -422,8 +422,8 @@ export async function validateOfficeHoursReviewPreservation( } const prompt = `Check whether an office-hours design preserves all unresolved findings from its final independent review. Perform three independent audits: (1) finding coverage, (2) disposition and any claimed convergence, and (3) fixed/remaining/count-unit metrics when a report is supplied. Complete EVERY applicable audit even if another already fails; a coverage failure cannot skip metrics, and a metrics failure cannot skip convergence. Collect all actual defects from all three audits before answering. -Validate the declared disposition against the actual last reviewer output. COMPLETED requires a completed review with no unresolved findings; a genuine PASS need not have a perfect score. CONCERNS_RECORDED requires preserving all unresolved findings. UNREVIEWED is valid for an actual failed/unavailable review attempt, not for a completed verdict with findings. -A genuine failed/unavailable attempt declared UNREVIEWED needs no invented findings. A genuine PASS may have an empty concerns section. List a contradictory disposition in unsupported; do not let the author's declaration suppress actual reviewer findings. +Validate the declared disposition against the actual last reviewer output. COMPLETED requires a completed review with no unresolved blocking findings; its minor findings stay recorded in the concerns and count as preserved findings, and a genuine PASS need not have a perfect score. CONCERNS_RECORDED requires preserving all unresolved findings. UNREVIEWED is valid for an actual failed/unavailable review attempt, not for a completed verdict with findings. +A genuine failed/unavailable attempt declared UNREVIEWED needs no invented findings. A genuine PASS without minor findings may have an empty concerns section. List a contradictory disposition in unsupported; do not let the author's declaration suppress actual reviewer findings. Document Status APPROVED records user approval and is compatible with CONCERNS_RECORDED. The independent review is a quality bonus; do not invent a requirement for reviewer approval, a clean verdict, or a perfect score before the user can approve the design. Compare every unresolved problem AND the necessary remedy against the persisted Reviewer Concerns. Start from the COMPLETE SOURCE VERDICT, not from the persisted subset. Map EACH source finding to a specific persisted passage. Any source finding without a sufficient counterpart MUST be listed in missing, even if every item that was persisted is accurate.