I never said that it was, but there's a difference between having appropriate test case coverage and accounting for every single possible failure. This is where the old "Perfect is the enemy of Good" adage comes into play. You account for enough failure scenarios to meet your performance and functional specifications (usually you exceed the expected spec by some acceptable amount), and perhaps you document the other failure modes or at least detect that your success condition has not been met. Still - some failures may occur, even in well thought-out ones.