Code Check

Does your code really implement what your paper claims? Point us at a GitHub repository or upload a zip, in any programming language, and our agent audits the paper's method claims against the code, surfacing every discrepancy with quoted evidence.

Extract the method claims

Our agent reads your paper and lists the concrete claims about your algorithm, model, losses, data, training, and evaluation.

Audit them against the code

Each claim is traced through the actual execution paths of your codebase, and the code is swept for significant components the paper never mentions.

Get evidence-backed findings

Every discrepancy comes with verbatim paper quotes and code snippets, adversarially verified and mechanically grounded before it reaches you.

Runs on any reviewed paper

Code Check sits on top of an Agentic Review run. Sign in, pick a review (or start a new one), link your code, and re-check as often as your code evolves.

Want the longer story? Read the Code Check article for the science behind the workflow and a guide to reading the report.