Comment by fzeroracer

How do you know they produce false negatives over false positives? How do you know if your top candidates simply didn't cheat over the honest candidates that produced worse but honest code?