paper-claim-audit

Paper claim auditing identifies every quantitative claim in a manuscript and extracts the claimed values. The audit matches these claims against exact data evidence from raw results to produce a report with per-claim statuses and a final verdict.

Zero-context verification is an auditing approach where the tool accesses only the manuscript claims and raw results without prior assumptions. It ensures every numeric claim is faithful to the underlying data, preventing confirmation bias and data misreporting.

You generate machine-readable artifacts by running the audit against the paper source and raw results. The process outputs a PAPER_CLAIM_AUDIT.json file containing structured per-claim statuses for downstream reproducibility checks and audits.

Traceability works by linking each extracted claim to exact evidence within the raw data files. The audit assigns a clear status—such as exact_match, rounding_ok, or mismatch—to every claim, ensuring fidelity in the verification report.

Yes, the zero-context workflow is designed to audit papers without prior knowledge of the study's assumptions. It strictly isolates the paper claims and raw results to verify quantitative data without introducing external confirmation bias.