The AI test that passes because it copied the code's assumption

A coding agent writes the function and the covering test in a single pass, CI turns green, and the reviewer merges. The problem: the test derived its expected value from the same logic as the implementation, so both carry the same wrong assumption and the test can never fail against it. This piece defines the mirror-test pattern, explains why code and test generated together share one blind spot, and gives a cheap mutation probe plus the characterization-test habit that tell a real assertion from a copy.