paper-claim-audit

To verify numeric claims in a LaTeX paper, you can audit every number, percentage, and comparison against raw result files. This process extracts statements directly from .tex sources and reconciles them with underlying .json, .csv, .tsv, or .yaml experiment outputs to catch mismatches.

Zero-context evidence auditing is a verification method that checks paper claims using a fresh reviewer thread with no prior interpretation or executor summaries. This approach reduces confirmation bias by examining raw metrics independently to ensure stated deltas match exact experimental values.

You audit rounding checks and arithmetic deltas by performing claim-by-claim reconciliation. This extracts every numeric statement from LaTeX and validates it against exact raw values, explicitly handling config mismatches and rounding discrepancies to ensure publication-ready accuracy.

Yes, automated claim extraction can check abstract metrics against raw .json and .csv files. It locates paper .tex sources and raw evidence files to perform zero-context verification, ensuring every stated evaluation scope and metric is fully traceable to its original experiment output.

The best way to generate an auditable JSON artifact for paper claim verification is to run an automated audit on your paper directory. This produces a structured report with hashes and trace metadata, alongside a human-readable Markdown summary with a PASS, WARN, or FAIL verdict.

A limitation of automated paper claim auditing is that it requires locating both paper .tex sources and raw evidence files like .json or .yaml to function. If raw experiment outputs are missing or config mismatches exist, the zero-context reconciliation cannot fully verify arithmetic deltas or evaluation scope.