What it is
Reviewer-side paper forensics that audits self-consistency and fabrication signals, then issues a deterministic integrity verdict from span-anchored evidence. Audit a paper for internal contradictions, weak baselines, citation issues, proof gaps, and reproducibility red flags, then produce a reviewer-ready report with deterministic verdicting from evidence-linked findings.
Key results
- 46 verdict-bearing patterns across 8 integrity families
- 13 AI-style signals are zero-weight and excluded from verdicts
- Deterministic core got 100% recall on 7 injected defects, 0 clean FPs
Shows paper-integrity auditing can be structured as evidence-anchored, deterministic forensics rather than opaque AI-text detection.