What I mean by "sample normalization" in this instance is that, put another way, the researcher (moonslav in this case) is adding a specific biased condition when doing his analysis that alters the nature of the result. That's even more significant when considering the boundary issue mentioned above, so even while accounting for that fact, there's some noise that would have to be eliminated through the use of linear weights, and just lol no, I'm not going into that. Point is, eliminating the worst five games doesn't actually (in my opinion) prove anything conclusive.