paper-claim-audit

To verify paper claims against raw evidence, use a zero-context audit that extracts every number, percentage, and comparison from your manuscript and traces it to raw metrics and config files. This ensures exact matching and flags rounding drift or aggregation mismatches.

Yes, a paper claim audit checks rounding and arithmetic integrity by flagging rounding drift, best-seed cherry-picking, delta miscalculation, and aggregation or run-count mismatches against declared raw result files. This applies to manuscript submission readiness and post-writing reconciliation.

A zero-context paper-to-evidence audit uses a fresh review thread to extract every quantitative claim and scope statement from your paper, then traces it to raw result files. This prevents confirmation bias by avoiding prior summaries and narratives during verification.

Yes, the audit requires strict input hygiene using your paper .tex sources and raw results or config artifacts. It excludes executor summaries and prior audit narratives to ensure numbers, rounding, configs, and run counts are reconciled against declared result files only.

The audit generates PAPER_CLAIM_AUDIT.md and PAPER_CLAIM_AUDIT.json files containing a PASS, WARN, or FAIL verdict. These outputs document whether every quantitative claim in your manuscript exactly matches the corresponding raw result evidence.

You should run a paper-to-evidence reconciliation check during manuscript submission readiness, post–paper-writing reconciliation, and before or after improvement loops that may change metrics, tables, or captions. This ensures every quantitative claim matches raw evidence.