paper-claim-audit

To verify numbers in a research paper match raw results, you need a zero-context claim audit that extracts every quantitative statement from .tex files and maps it directly to raw result files. This process outputs a per-claim reconciliation report with PASS, WARN, or FAIL verdicts.

Zero-context review prevents confirmation bias by auditing paper claims without using prior logs or summaries. The reviewer evaluates only the paper .tex inputs and raw result files independently, ensuring that text interpretation does not drift from actual experimental outputs.

Detecting best-seed cherry-picking and rounding drift requires comparing reported paper numbers against raw result files claim by claim. An automated audit flags these common integrity failures alongside config mismatches, aggregation errors, and delta arithmetic mistakes.

Yes, paper claim auditing works directly with LaTeX .tex files and raw result evidence files. The process reads only these inputs to extract quantitative and scope claims, mapping them to exact evidence values to produce machine-readable and Markdown audit reports.

You should run a paper verification audit right after your paper draft is produced and again before final submission. This ensures that every reported number, comparison, and scope claim remains backed by raw results after text edits and improvement loops.

A paper claim audit detects common integrity failures including rounding drift, best-seed cherry-picking, configuration mismatches, aggregation errors, delta arithmetic mistakes, caption and figure misalignment, and overclaimed evaluation scope by tracing claims to raw evidence.