What problem does it solve? Scientific manuscripts often contain claims that drift from the underlying experimental data, leading to hallucinated percentages, fabricated results, or unsupported conclusions. This Skill audits a LaTeX paper against its raw results file to catch such discrepancies before publication. ## Core Features & Use Cases - Claim-to-Data Auditing: Calls the audit_scientific_claims MCP tool to compare every claim in paper.tex against the raw results.tsv data. - Structured Validation Output: Writes a machine-readable validation summary to workspace/rigor_audit.json for downstream review. - Manuscript Rejection Loop: If the audit fails, it rejects the manuscript and instructs the write-worker to correct discrepancies. - Use Case: In an autonomous ML research pipeline, after a write-worker drafts a paper from experiment results, run this audit to verify no fabricated percentages or false claims slipped into the manuscript. ## Quick Start Audit the claims in workspace/paper.tex against workspace/results.tsv and write the validation summary to workspace/rigor_audit.json.
