TalkChess.com

[Mike Scheidl]

This actually supports my doubts. The method spots eval similarities but does not spot, to a sufficient degree, search differences. 100ms vs. 400ms still give ~73%. That means, if someone takes Fruit 2.1's evaluation but makes the search code four times more effective (leading to your 400ms results in 100ms), this achievement would not be recognized sufficiently by the testing method.

This is a problem.

I insist that a similarity analysis must not ignore differences of the search code. It cannot be based on eval only, because reasonable eval terms are within a limited horizon. It is no surprise that they are similar among very good engines and only small details are different.
_________________
Regards, Mike