I agree it looks similar to QuickCheck shrinking. Isn't necessary limited to compiler writers though; e.g. I can imagine a big numerical calculation (like a phy

I agree it looks similar to QuickCheck shrinking. Isn't necessary limited to compiler writers though; e.g. I can imagine a big numerical calculation (like a physics simulation), giving NaN outputs. This could presumably home-in on the problematic step (given an appropriate test script).