It sounds almost like it is a kind of fuzz-testing then; I'm interested in what percentage of problems uncovered were problems in the spec or undefined behavior

It sounds almost like it is a kind of fuzz-testing then; I'm interested in what percentage of problems uncovered were problems in the spec or undefined behavior, versus problems introduced by the simulated annealing itself.