paper-claim-audit

Paper claim verification works by cross-referencing text against raw experimental result files to detect discrepancies in metrics, rounding, and configuration. It requires a zero-context model execution to ensure independent validation of research fidelity.

Checking for data integrity issues involves auditing paper source files against raw data outputs to identify failure modes like number inflation, cherry-picked seeds, config mismatches, and aggregation errors before submission.

Yes, you can audit experimental evidence stored in JSON, CSV, and TSV result files for rounding errors. Evidence verification automatically cross-references these raw formats against paper claims to detect mathematical discrepancies.

Zero-context verification eliminates confirmation bias by executing a fresh, independent model run for validation. This strict audit ensures every quantitative claim is precisely supported by raw experimental evidence without prior context.

The best way to detect config mismatches is running a strict evidence verification audit on the paper directory. This compares text claims against raw experimental result files to identify configuration errors and ensure mathematical accuracy.

You need to run a paper claim audit before submitting a research paper to verify all quantitative claims. This ensures reported metrics, percentages, and comparisons are mathematically accurate and fully supported by underlying experiment logs.