More OTB Cheating !

Common, be serious! Nobody knows of Dembo and Docx. Their accounts are not shown as being closed for cheating. There is no reference, nothing! It just didn't happen! You dreamed.

About the case of Dembo - I think the problem is that titled players are often the target of cheats, so how can you play a game when your opponent is cheating?

Anonymity - I think perhaps sometimes the trash is the player who constantly trolls and ruins people on the internet until they have to make a new account.

@Fezzik: Another problem I already noted on the Cheating forum was that perhaps some players cheat on some games - i.e. against a very low rated opponent because they do not want their rating to drop. The stats for individual games are also important. I find that checking over a game with an AI kibitzer is usually interesting, If the AI is at 20 ply and one player makes 95% of the computers 1st choice then I would be very suspicious.

Touche!

This is a good point and is why many titled players choose not to play online. This is  also why, imo, those who brag/boast of beting a titled player online are making empty/meaningless boasts. If you havent beaten a titled player otb, face to face, then you havent beaten a titled player imo. 

I was responding to a particular post ( which I quoted ) , which also had nothing to do with OTB cheating and thus, neither did my response. 

I think that the results of the analysis were mentioned in this thread. Here is the link, again:

http://www.chessvibes.com/reports/french-cheating-disciplinary-committee-says-guilty/

I really doubt you can perform the test yourself. It looks to me that you cannot understand it at all. A fundamental principle of counting move match-up is that different engines, running on different computers, will produce almost the same results.

Yes & no.

The likelihood that the match up % results would be very similar on different systems using different engines (albeit one of the 3000+ Elo rated ones) depends largely on a decent amount of non-database moves from plenty of games in the sample.

I got

{ Feller, Sebastien (Games: 9) }
{ Top 1 Match: 186/264 ( 70.5% )
{ Top 2 Match: 224/264 ( 84.8% )
{ Top 3 Match: 237/264 ( 89.8% ) 
{ Top 4 Match: 247/264 ( 93.6% ) 

Using Houdini 1.5 x64 Hash:640 Time:45s Fixed Depth:20ply

on a quad-core, but there are only 264 moves analysed, which is about one-third the size that this analysis typically deals with.

The method was not designed for small samples & the chances of a false positive (or negative) and variance for results from different sources increases as the sample size decreases.

The results above are the ones from Khanty-Mansiysk 2010 Olympiad.

What games did you want analysed as a comparison?  If you posted the pgn (or PM'd them to me) it might be helpful but as I say, small samples are not what this analysis was intended for & results would have far less meaning than for a large sample of 20 games with let's say around 750 non-db moves.

Apparently the cheating happened only in four games:

Update: we received Fressinet’s conclusions from analysing four games (Howell + the last three, against Timofeev, Gelashvili and Efimenko):

• against Gelashvili and Efimenko, always 1st choice of Firebird
• against Howell, same except for move repetition on move 37
• against Timofeev, once 2nd choice, once 3rd choice, all the other times 1st choice until the position was -5.

Steve, do you have to do the analysis and move comparison by hand or is the process automated in some fashion?

Do you just set the engine to Full Analysis with the parameters noted or do you have to walk through using Infinite Analysis and compare?

Steve and I use the same software, Batch Analyzer. You load the pgn and then just sit back until the results are in, after setting time per move and ply depth.

Depending on the exact methodology, one may do some manual analysis as a final step. I do this in Chessbase.

It's not such a great pain though compared to the fully manual method which Steve used for hundreds of games before BA. That took exceptional perseverence.

Ah! Much more efficient. Is batch analyzer available for download or purchase?  I did some google searches for it and came up empty.

I did, however, find some cool papers on hash and search by the guy that wrote Crafty: http://www.cis.uab.edu/hyatt/pubs.html

EDIT: After more research I found this program: http://www.chess.com/download/view/chessanalyse-25

Will it do the same thing? I found a blog by this guy who says he used it: http://polarbearspalaver.blogspot.com/2011/04/methodology-and-statistics-used-in.html  but I can't tell if it permits batch analysis (without experimenting, of course, just want to avoid that if someone has an easy answer).

Thanks!

jhb

BA was written by one of the games mods at RedHotPawn, as a tool to detect engine users there. As such he has insisted that his program be in the hands of the few and hopefully trusted, since if it became widely available to cheaters they may be better armed to avoid detection.

I have heard that Aquarium has a scriptable feature that can analyze batches automatically in a Top 3 method but I've never used it myself.

A member here (gwhuebner) has created a program (ChessAnalyze) that can also be used for batch (multi-game) analysis.

Apart from that I don't know of other software that can do this kind of analysis.

To get the most reliable results one must establish benchmarks using the same hardware, software, and settings that will be used in the analysis of the suspect games.

Despite the automatic aspect of using such software it's not a trivial undertaking to establish such benchmarks. In my case I implement an opening book using Megabase and a correspondence database, and this is by-hand work.

By my count just four members have done serious benchmarking work: Steve, Polar_Bear, cludi, and me.

This doesn't mean that analysis sans benchmarking is worthless; it can be a fairly reliable way to sort out suspect engine users.

When your settings and hardware aren't utterly different from the benchmarking setup, you can lean on those benchmarks some. Often a handful of us have simultaneously analyzed a suspect, and the agreement between those who have benchmarked and those who have not has been remarkable nevertheless, especially on the 3rd match up number.

I see those who run a game through Fritz on unknown settings and claim high matchups and then lay the accusation of engine use, but this is a very rough way to run analysis.

Old otb greats have managed 100% Top 3 3rd number match ups for single games and they certainly weren't using engines.

For this reason, as Steve has mentioned before, we try to use a 20 game batch and c. 700 moves (most recent vs. "strong" opposition).

That's the classic Top 3 method. We've developed other wrinkles, but that's classic T3.

I said you ought to show some work first, which we've never seen. We've seen enough of those who have grandiose ideas of how things should be done re cheating detection but are in fact at sea when it comes to undertanding the topic.

If you want to use Gerd's program, feel free, and show your benchmarking results.

edit: I don't know why you'd think it some kind of honor to work with you on the project of your choosing in the topic of engine use detection. You don't have any bona fides on the topic and you've often been bumptiously in error. Got a good idea you want others to work on with you? Demonstrate it first!

Maybe someone is using a magic wristwatch, ring or earring ...

You understand nothing! Absolutely nothing! One can do a reliable test by hand. One game takes at most one hour.

Now here is what should happen in the future, should Feller, Hauchard or Marzolo ever participate in a chess tournament again.

• The organizers and all the other players set up a secret lab with Rybka and Houdini software, to aid whoever is scheduled to play the bad guys
• Every time one of the three plays, the lab room is consulted, and best moves recommended by the software are ostentatiously passed on to the opponents, sometimes via little paper notes, sometimes whispered into the player's ear by a passer-by
• After the bad guys lose in 25 moves to folks rated 1733 and lower, they get suspicious: "Monsieur, I believe I 'ave witnessed cheating in this 'ouse", says the cheater to the TD. "No, Sir, I assure you we have implemented the strictest safeguards!", responds the TD
• The above goes on in every tournament they participate in, until, hopefully, they quit playing altogether.

The board could contain lights that can only be seen from the angle of their opponents, the computer lights up the squares to move to ... or even laser pens ... they could have the lasers hit the side of the piece that the cheaters opponents are using.

I'm surprised cheaters aren't immediately apprehended and taken to jail for stealing.   

Reading this post reminds me of the movie '21' about a team of gamblers with an elaborate scheme to beat the casinos out of millions of dollars in the game of blackjack.  I had no idea this went on in the chess world.

Yes but Dan decided to start with Capablanca-Alekhine 1927 as his first benchmark.

This was about 6 months ago & he stated that the top 3 analysis was to be done at fixed depth = 20 ply.

I analysed the 22 games from this match which each had 20 or more non-chesslive.de moves (as of 1927) and there was a sample of 852/853 non-database moves.

Since even with Batch Analyzer, a fixed depth=20 analysis of a single game can take 4 hours I think Danny boy may be some time!

He may also have discovered why a time cut off of 30 seconds per ply was used for my analysis of this match.  Analysing this match by hand (before I had Batch Analyzer & doing around other commitments) took me about 2 weeks in June 2009.