I also think other factors could objectively be used to exclude certain kinds of games. One such factor is the opponent's own fidelity. <kingkoy301> scored a remarkable 31/33 top 1 choices in his game against her, and he is accepted as an honest player. Since you can't suspect one player and not the other, you'd be forced to assume there is something about the game itself that created the high fidelity. You would then need to discard it to prevent unfairly skewing the results.

<cotoi> kingkoy301 is a member of good standing. He's playing now: http://www.chess.com/echess/game.ht...

Their game was in May, yet he's survived the scrutiny of Chess.com's awesome team. The analyzers can't suspect her percent, while accepting his. Their game must be dismissed, leaving 19 games, short 1 game to have launched an investigation. <Zygalski> You should have noticed these kinds of games when you were reporting Yelena's results to the Chess.com authorities.

The game shows the weakness in the Top 3 method when it is used so crudely as Chess.com does. Top 3 needs to be a part of Chess, rather than a numbers game.

On forcing moves you'd set up an increment giving more power to smaller differences of evaluation between candidates multiplied by the number of candidates within a threshold.

<You might have a tactical game, with many recaptures and only-moves and you obtain a high top-choice agreement. Yet, it's unlikely to get a very high agreement with all the top-3 choices.>

What? NO! If you get a high top choice agreement, then all the rest of your choices are at least as high as the top choice. kingkoy301's 31/33 leaves virtually no room for improvement. Besides no one is suggesting that highly tactical games are easy to play. There are games which contain lots of recaptures because one of the strategies you should employ after gaining an advantage is to force exchanges.

<Counting forced moves will not skew the results since the same approach was used for the games in the benchmark.>

You are now contradicting your own opinion about kingkoy301's game. The benchmark for cc was almost passed by Yelena in her OTB game, it is just too low and a poor normal control:

Yelena's Olympiad............64....72.......76%
Heritage benchmark..........65....80......90%

A single game is not enough proof of anything, but it should show that chess.com's method should be more rigorously tested.>>

That, of course,is rather subjective. But in most cases it is easy for s player to explain why a move is made, but I find many of Rybka's moves baffling.

All that said,I do have a suggestion for presenting the human/machine comparison in graph form. Put a data point at the top of the graph every time the computer's best move is matched, 25% lower for second, 50% lower for 3rd and 75%n lower for 4th. This makes it easy to follow and probably can be automated using a spreadsheet with graphing capabilities. A picture is worth 1000 words (or numbers).

Think of each game as a datapoint, and the entire 20 games as a dataset. In statistics, datapoints will be rejected as outliers if they are located far outside normal values. It's important to identify the outliers if you want to understand the truth about what's being measured.

I & others too have said that a few (or a single) cherry-picked game(s) mean nothing in terms of this methodology. The key factor is the match up rates in many games over time.

It doesn't matter if Dembo matches top 3: 100% in an otb game, as I said, one of the analysts already looked at Dembo's otb:

"I picked the last 20 games >= 35 moves from chessgames.com. All against high-rated opposition (lowest ELO in the set is 2222, highest is 2655)"

thinking that perhaps Dembo's natural style of unassisted play is rather more engine-like than should be expected. The results of the analysis were:

The analysis process alone takes some time. Maybe chess.com doesn't have an army of highly-paid, titled staff all analysing 100's of games constantly?

Would you rather they banned members in a haphazard, hurried fashion or that they prioritize & ban the most blatant cheats first?

I selected Yelena for batch analysis because
1)she had 10-15 games in progress vs 12 players rated a chess.com average of 2614. She was moving in a daily basis in all games, as I recall. 2)her win/loss/draw record of 140 wins/0 losses/15 draws. 3)she was #21 highest rated and climbing, which meant she was well within the top 0.1% highest rated on site. Most engine users are booted from the top 1% (2500+ rateds) so there was a likelihood that she would be playing games against some other engine users whilst maintaining her unbeaten record.

This can al sound like nonsense to you - as you wish.

The simple fact is that the analysis for these Dembo games suggested blatant engine use, from 4 separate analysts. This evidence was forwarded to chess.com staff for them to run their own checks (whatever they are!) and Yelena was later removed from site.

<@tpstar>: the analysis is not retrograde. It's a big difference between going backwards (basically asking the engine to check if some moves are good) and going forward (asking the engine to discover the best moves). The OTB versus CC dilemma has been addressed repeatedly.

There existed about a 300 point rating difference over 19 of Yelena's 20 games, with an 8 point rating difference in the most recent game. There was also at least a 400 point strength difference from the 9 games taken from 2008, with opponents averaging 2250-Chess.com (FIDE: 2050) while her FIDE rating was 2450. We know that the less strong is the opponent, the easier it will be to find good moves, and the higher fidelity will be achieved. Yelena demonstrated this principle in two of her remarkable Olympiad-2010 games.

This suggests her high-fidelity are the result of rating differences. So it is critical to have used the proper benchmark with which to measure such a batch. At the very least, your benchmark-batch should have been derived from correspondence games that have opponents so disequally-matched - as any qualified statistician would know.

This suggests that you aren't a qualified statistician, aren't qualified to defend this benchmark, and certainly aren't qualified to be convincing others of your flawed-design. By what expertise can you even suggest she is guilty of "blatant engine use".

To single out individuals for cheating is pointless and stupid. While there is internet chess, this problem will exist forever.

Chess on the internet should be for fun. If some 1600 Elo rated player wants to start an account with an internet chess site, and they decide to use Rybka at the start and get a rating of 3200, then he's stuck with having to play computer chess for his duration on the site. They will soon realise there's no fun in that.

But if a person changes and decides to be honest, then there's always someone to replace them who's just as dishonest. Round in a circle we go.

Would you play correspondence chess for money if the rules say 'no computers.' I definately wouldn't.

The site owners are kidding themselfs that they can find and control the cheats.

Some cheat by picking computer best moves, some cheat by mixing the best move with the 4th best move, some don't pick computer moves but check their analysis with a computer.

Some may not cheat at all and then suddenly they have to go to a relatives wedding and they don't have time to analyse, so on just this one occasion they let the computer show them the way.

How do you control who's doing what. If you ban Yelena, then there must be hundreds more who at some level have had a snidy look to see what the computer says.

Too much controversy. Allow computers in correspondence chess and be done with the flame wars and character assassinations.

Never take internet chess seriously, it could damage your health. You never know when it's a true game or not.

I don't agree with your numbers: I think there are thousands of cheaters, not hundreds.