Gating agent-written patches on fixture sensitivity instead of green test suites

A passing test suite can be weak evidence for an agent-authored patch, since tests may pass because they have stopped constraining behavior. The article proposes measuring how many deliberate breakages a fixture still catches, rather than only whether tests pass. Its approach combines property obligations, canary mutations, and a freeze registry with expiry, wired into a pre-merge check. The main risk described is fixture drift, where fixtures and assertions gradually loosen as patches are added. The gate is designed to detect this loss of test strength without assuming malicious intent.