That's kind of brilliant, although I could see it being a clusterfuck in actual implementation, 30 isn't enough per-test to get meaningful statistics for one th

That's kind of brilliant, although I could see it being a clusterfuck in actual implementation, 30 isn't enough per-test to get meaningful statistics for one thing. It also involves a total lack of control from the top-down, so it'd probably be resisted pretty strongly. I really like the basic idea, though.