Test Coverage and Metric Integrity

When the agent owns the denominator, "all tests pass" stops being a measurement

A dev.to tip argues the failing test must exist before the fix does, and that any drop in test count should fail the build. The arithmetic is on its side.