Paul McKeown wrote: ↑Thu Apr 01, 2021 8:08 am
I have a question for those that know something about the statistical matching methods used to compare players' moves with engines' best n moves.
Is there some allowance made for opening preparation being made in events which have moved from otb to online, particularly where long time controls have been retained?
All experienced players will check their opponent's repertoire before the game and make better informed choices as to what openings and variations of those openings they should make than if they had not prepared. So a decent club player may in the hour before the game prepare for the game and play the opening score of moves as well as a titled player would, before inevitably subsiding into their normal slightly substandard sludge of middlegame and endgame play.
However, this seems different to me from the normal way that online chess happens, where players play scads of games without any preparation at all.
Is there a risk that some of the scandals of the past year are the result of statistical methods picking up this directed preparation by experienced players, rather than actual cheating?
For chess.com a ban for any titled player would have to be signed off by three GMs and one of the considerations they would be looking at is the possibility of opening preparation. That is pretty accepted practice, so high profile cases are going to have some level of 'expert' input. Obviously, there simply isn't the level of resources to ensure that all cases receive this level of scrutiny.
So could a decent club player trip the cheat detection with repeated targeted preparation, well yes, but it is very unlikely. For the platforms, they would show up as playing well, but there would be nothing else suspicious. They don't know that The user John Smith is actually John Smith graded 150, it could be that he is David Howell or Magnus Carlsen for that matter
When the Regan tests are performed then they do know that John Smith is John Smith. The test comes in two parts the initial test, which simply looks at moves matches and then the comprehensive test which produces a Z-score. Our ardent preparer is going to look bad on the initial test because they are going to have a higher than expected move match.
When the the full test is performed all theoretical moves ( as defined by moves played by 2300+ players) are removed, so things should look a lot better for John.
Of course, if John has done lots of computer analysis and expanded on existing knowledge, remembering lots of lines with all his opponent's responses then it is possible that he'll fall foul of both Regan tests, but in reality how likely is this to be the case for a decent club player?