What problem does it solve? Reproducing scientific papers often fails silently because dataset versions are unpinned, inline thebibliography entries drift out of sync with .bib files, and verification steps are skipped without records. This Skill audits a paper's working directory against a golden test set to determine whether its results are actually reproducible. ## Core Features & Use Cases - Dataset Version Verification: Checks that dataset versions are locked (e.g., via a lock file) before accepting any reproduction claim, blocking unpinned data as "reproduction without evidence". - Dual Bibliography Sync Check: Detects drift between inline thebibliography environments and .bib files, requiring both sources to be updated together. - Golden Set Testing: Runs weighted pass/fail checks (critical/high/medium) against golden cases, requiring a weighted score of at least 0.80 with all critical checks passing. - Failure Traceability: Requires every verification step to be executable, recorded, and reproducible, with failures logged alongside concrete recovery actions. - Use Case: Given a paper directory with a pinned dataset, 32 bibliography entries in both sources, and a runnable pipeline, produce a reproducibility report confirming version consistency, bibliography sync, and golden test passage. ## Quick Start Audit the paper directory 'papers/repro-target' for reproducibility and generate a report covering dataset version locking, bibliography synchronization, and golden test results.
