A single-file Python CLI that hashes your ML accuracy claim with SHA-256 before the experiment runs. Edit the threshold afterwards → CI exits 3, the lie is mechanically blocked. MIT, on PyPI: pip install falsify.