What problem does it solve? Researchers and reviewers often need to verify whether a published paper's methods, defaults, and metrics actually match its released code. Manually cross-checking claims against implementation is slow and error-prone, and README files alone rarely reveal real behavior. ## Core Features & Use Cases - Claim-to-Code Comparison: Reads the implementation, tests, configuration, and history to compare the paper's methods, defaults, metrics, and data handling with the actual code. - Reproducibility Risk Detection: Reports missing implementations, ambiguous defaults, unstated seeds or RNG kinds, and reproduction risks. - Structured Audit Report: Produces a single report at docs/<slug>-audit.md ending with a Sources section of direct paper and repository URLs, optionally delegating evidence gathering to researcher and verifier agents. - Use Case: Before citing or building on a paper, ask for an audit of the paper against its GitHub repository to confirm the reported metrics and default hyperparameters match the released implementation. ## Quick Start Audit this paper against its public repository and write a report of any mismatches between its claims and the implementation.
