Originally posted by sonhouse
How many of his games were included in the analysis? Could the analysis show games that were totally human, like was there a pattern, say 10% computer match for a couple of games, then 80% for one game then back to 20 for a game, etc.?
Can they get that kind of detail?
It's reasonably to include a large sample, say 20-30 games
I doubt that in the case of Weyerstrass there is a game one can say "that's clearly a human game!". When I first came to this site, I randomly picked two of his games and used an engine to see how the site makes sure that cyborgs are not allowed to play here. Needless to say, in both games Weyerstrass matched Rybka consistently.
There are more things you can use to detect cheaters even by analyzing a small number of games: tactical errors and total error in a game. Usually, in a complicated middle-game position between two strong human players, the evaluation of the engine fluctuates let's say from -0.50 to 0.50. It means nothing, of course, it's just how the engine says that a position is equal. Well, things were different with Weyerstrass: even if, for a certain move, there were like 6-7 good candidate moves, he always picked one considered optimal or near optimal (within 0.20).
Your numbers with 10% computer match are totally wrong! Even an 1600 player will match the top choice of an engine like 40-50%.