See Pete Lindstrom’s Spire Security Viewpoint for empirical evidence that mechanical quantitative diagnosis is almost always at least as good as clinical qualitative diagnosis.

There is still plenty of room for qualitative decision-making in arenas where there aren’t enough facts or the facts haven’t been quantified or there’s no baseline or there’s no mechanical method yet. But where those things are available, it’s better to use them. You’ll still need qualitative judgement for cases where the algorithm is right but it didn’t take into effect unfortunate side effects, for instance. Even then, you’ve got a better chance of knowing what you’re doing.

-jsq