Chess Player Strip Searched

[KWRegan]

Roger, count me among your "anyone"---the games in my datasets come from various sources. E.g. OpeningMaster has many more games by sub-2000 players than ChessBase Big not to mention TWIC, and I am grateful for efforts like CANbase and would like to know of more.

In cases like the recent incident with a 2230-player at the Iasi Open, where I'm naturally curious to try detecting indications in the games, it is reasonable to limit to a broadly comparable population of events. Thus sub-2000 sections and weekenders wouldn't be in the relevant population to begin with. (Incidentally, Mark C. has all games I can find on the Iasi official site, and my other contacts have not yet been able to procure more.)

I don't understand the sentence beginning "It's all very well...", but please reference my sentence above about screening vs. full tests and when full tests are called, which is also inherent in the document. Regarding your reply to MartinCarpenter, the document's criteria for the full test in Appendix C prescribe: "The tests should measure specific criteria, such as (...), so that a positive result has more specific meaning than `this person played unbelievably well'.” My "Crown Game Affair" article says "Thus I claim specific value for my tests beyond being a metric of performance", while my linked open letter (http://www.cse.buffalo.edu/~regan/chess ... report.pdf) goes into details, lays out the procedure end of the "recipe", and also addresses other concerns voiced in this thread. If you are dissatisfied with the answers, please reference them and say concretely what better diligence needs to be done.

[Roger de Coverly]

I don't understand the sentence beginning "It's all very well...", but please reference my sentence above about screening vs. full tests and when full tests are called, which is also inherent in the document.

It's my understanding of the proposal that all rated games should potentially be screened. If any are found suspicious because the player consistently outperformed their rating with their move selections, then the next stage of the investigation takes place. The initial stage is independent of any external evidence.

Contrasting this with playing on an on-line server where there are no witnesses, a lower standard of suspicion is allowable.

Is it, or is it not the case, that the proposal would allow a player to be banned <b>purely</b> on statistical evidence? British readers may be reminded of this case which shows the dangers of relying on misinterpreted statistics and dismissal of one in a million chances.
http://en.wikipedia.org/wiki/Sally_Clark
http://en.wikipedia.org/wiki/Sally_Clark

[Paolo Casaschi]

1. The 20-year estimate for a 1M-to-1 false positive (using 4.75 sigma as criterion) is based on about 200,000 tested games per year

Why only 200k games/year?

The last FIDE rating list reports 38k players with rated games during the month. Assuming each player played on average one event, the it would be 38k events/month, about one million events every two years hence a risk of 1 false positive every two years.
After all the ACC recommends every event to submit games and you probably will attract all of the "suspect" ones where the false positive could hide.

Not sure you answered this one already: would you propose to make public the full details of the testing procedure (up to the point that anyone could build his own replica) or would you suggest to keep the details of the testing procedure secret (in order to avoid cheaters to work around the system?
In my opinion, in order to safeguard the legitimacy of the process, the former is required. I find difficult to accept disqualifying a player for several years based only on a test that can not be openly replicated.

PS: welcome to the ECF forum; you will find a lot of opinionated people with more time available for wasting that knowledge to share, but you might came across the occasional helpful and passionate chess player... You have been warned

[Paolo Casaschi]

1. The 20-year estimate for a 1M-to-1 false positive (using 4.75 sigma as criterion) is based on about 200,000 tested games per year

Why only 200k games/year?

The last FIDE rating list reports 38k players with rated games during the month. Assuming each player played on average one event, the it would be 38k events/month, about one million events every two years hence a risk of 1 false positive every two years.
After all the ACC recommends every event to submit games and you probably will attract all of the "suspect" ones where the false positive could hide.

Not sure you answered this one already: would you propose to make public the full details of the testing procedure (up to the point that anyone could build his own replica) or would you suggest to keep the details of the testing procedure secret (in order to avoid cheaters to work around the system?
In my opinion, in order to safeguard the legitimacy of the process, the former is required. I find difficult to accept disqualifying a player for several years based only on a test that can not be openly replicated.

PS: welcome to the ECF forum; you will find a lot of opinionated people with more time available for wasting that knowledge to share, but you might came across the occasional helpful and passionate chess player... You have been warned

Looked (quickly, I admit) at the PDF from the FIDE congress that should describe the recipe, but I'm still missing your definition of "matching". In other words, how do you go from a gamescore to the matching percentage?

[Roger de Coverly]

Not sure you answered this one already: would you propose to make public the full details of the testing procedure (up to the point that anyone could build his own replica) or would you suggest to keep the details of the testing procedure secret (in order to avoid cheaters to work around the system?
In my opinion, in order to safeguard the legitimacy of the process, the former is required. I find difficult to accept disqualifying a player for several years based only on a test that can not be openly replicated.

The other side of the testing is that there's a reasonable back catalogue of games where it is known that one or both players was a computer. There's also the back catalogue of Correspondence games where a certain amount of computer assistance is likely. I'm not expecting a detailed answer as it would show which engines don't play like engines, but how much testing has been done that the proposed method can detect engine use where engine use is already known? Kasparov famously accused the Deep Blue team of having human intervention. Particularly with earlier events when engines were weaker, can it be proved that all supposed engine v engine encounters were that and not human influenced?

If the premise is that you can detect cheating because a player of rating x wouldn't be capable of finding moves y, that places far too much faith in the reliability and stability of FIDE Elo ratings.

I can understand the frustration at the inability to nail you know who, but if you cannot establish the mechanism of the consultation, then with games taking place in full public view, you have to give the benefit of the doubt.

If it's not allowed or very risky to have even switched off mobile phone in your possession for any type of FIDE rated event, then one of three things happens:-
No-one has a mobile phone in their possession or
No-one enters tournaments or leagues under those conditions or
Tournaments and leagues continue, but outside the scope of FIDE regulations.

[Roger de Coverly]

in other words, how do you go from a gamescore to the matching percentage?

I'm not sure if I've understood it correctly, but it's more sophisticated than just saying that moves match the choice of an engine. The underlying premise seem to be that you can only play accurate moves ( as measured by an engine) unless you have a very high rating or are consulting an engine at every move. This has an air of IM/GM arrogance which is that if you aren't a member of their club and have the temerity to beat them or draw with them, then you have to have been cheating in some manner. Some GMs are more vocal about this than others.

I would say it's in the nature of chess strength that players below CM (2200) standard can be strong in some areas and weak in others, so their rating can reflect an average of their good performances and their bad ones.

[Paolo Casaschi]

1. The 20-year estimate for a 1M-to-1 false positive (using 4.75 sigma as criterion) is based on about 200,000 tested games per year

Why only 200k games/year?

The last FIDE rating list reports 38k players with rated games during the month. Assuming each player played on average one event, the it would be 38k events/month, about one million events every two years hence a risk of 1 false positive every two years.
After all the ACC recommends every event to submit games and you probably will attract all of the "suspect" ones where the false positive could hide.

Not sure you answered this one already: would you propose to make public the full details of the testing procedure (up to the point that anyone could build his own replica) or would you suggest to keep the details of the testing procedure secret (in order to avoid cheaters to work around the system?
In my opinion, in order to safeguard the legitimacy of the process, the former is required. I find difficult to accept disqualifying a player for several years based only on a test that can not be openly replicated.

PS: welcome to the ECF forum; you will find a lot of opinionated people with more time available for wasting that knowledge to share, but you might came across the occasional helpful and passionate chess player... You have been warned

Looked (quickly, I admit) at the PDF from the FIDE congress that should describe the recipe, but I'm still missing your definition of "matching". In other words, how do you go from a gamescore to the matching percentage?

Reading also the other PDF with the calculation examples for the Ivanovo case.
What I find surprising is the amount of arbitrarily chosen setting (running program x arbitrarily chosen among many and running at fixed depth Y again arbitrarily chosen). Amazing that the analysis should not be fundamentally affected by the choice of X and Y that in fact heavily determine the move chosen by the engine.
I need to do more reading necessary apparently....

[KWRegan]

Apologies again for not quoting individually while keeping one-on-many replies compact: The FIDE/ACP document does not fix a recipe---it lays out criteria one must satisfy. I claim mine does so, but even in my Tallinn talk I acknowledged that someone else could come up with a better one, even saying then I would "tip my King" . The in-progress upgrade will use values from all depths up to "X", while the current X-es and Y-s have stayed consistent for empirical testing.

Regarding 38k players/month, I would love to see that many PGN games on FIDE's site. Our intent was to place maximum discretion in arbiter and TD hands, with care as to "obligation" vs. "recommendation" in the three designated protection levels. The middle level says "obligation" only for norm-related games; providing them is already on the books. The "Standard" level says only "recommendation to send in games for screening". Maybe there's an inconsistency between (*) the idea that our planned web server is to be operated directly by TD staff for convenience and (**) the words "send in" sounding like ACC would do it. I noticed this only now, while the larger feeling as lovers of chess (as for Hugh Brodie with CANbase) is that we'd like to see more games sent in and preserved overall. We do wish arbiters to screen round-by-round so as to be better informed in advance if-and-when situations come up---in quite a number of "whisper-cases" the screening test alone has been enough to counter the complainer's own move-based claims. The document says on pp8-9:

The results of the “fast test” are to be kept confidential and are only meant to assist the Chief Arbiter in identifying cases that may call for further measures to assure that players are adhering to the rules. If requested, the ACC shall provide assistance to the Chief Arbiter in determining such measures.

We didn't say and didn't have in mind that the screening-test results must be shared with ACC---nor in Appendix C. If the arbiter perceives something and so requests assistance, that's another story. Otherwise our attitude was and is, "what happens in Vigo stays in Vigo." I hope this is enough to indicate we're not creating a structure where processing 500K player-events/year is centrally purposed or the attendant sampling considerations even emerge, also per my note about comparable populations above. If they do, and we get 2M not 200K relevant games per year, hurrah---and then the provision for "periodic review" of procedures comes in to play. 10M-to-1 is almost exactly z = 5.20, so we could change it---while I maintain 4.75 based on "TWIC with about 3x slack" (and 2.75 rather than 3.00 for cases with physical/observational evidence) is right for now with some room to grow. By then we might have at least a 10x-larger comparison base as scientific control, which is among several safeguards of the kind that were apparently lacking in the Sally Clark and "Dutch Nurse" case (another in the latter was omission of Fisher's or Stouffer's Rule for combining unlikelihoods).

Finally, touché on being unable to "nail" you-know-who, but at least I've acted out the hypothesis of it being within a hobbyist's reach to program a (hidden) camera to decode positions visually and send them to Houdini: http://www.chesstalk.info/forum/showthr ... n-Open-run Hopefully to be re-enacted at AAAI 2014 in Quebec City, along with presentation at a satellite workshop of the latest paper http://www.cse.buffalo.edu/~regan/paper ... 14aaai.pdf, which includes calibrating engines to each other.

[Roger de Coverly]

Finally, touché on being unable to "nail" you-know-who, but at least I've acted out the hypothesis of it being within a hobbyist's reach to program a (hidden) camera to decode positions visually and send them to Houdini

This thread was started with initial suspicions. Much speculation followed as to possible methods.

It has been suggested that a vibrating device in a shoe or elsewhere could be used to communicate suggested moves, which leaves open the question as to how the engine finds out the position. If it is plausible to use a hidden miniature camera to parse a board either for moves or current position, that adds another potential piece to the jigsaw.

No-one has yet tried this approach as a third way of broadcasting live games. The other two being sensory boards such as DGT and "key the move in" devices such as Monroi. The latter never really caught on in Europe, it being obvious even ten years ago that you really didn't want live electronic devices under player control anywhere near the board. The casualty of the legalisation by FIDE of the Monroi device was that the traditional blunder check approach, writing the move before you played it, had to be outlawed.

[NickFaulks]

The casualty of the legalisation by FIDE of the Monroi device was that the traditional blunder check approach, writing the move before you played it, had to be outlawed.

This is gibberish. The ban on writing a move before playing it, which destroyed a generation of training, had nothing to do with Monroi. A few arbiters reported that they had seen players writing a succession of moves on their scoresheet and scratching them out, which looked like making notes. They were too feeble to use their power to tell the player to desist and instead, in traditional FIDE fashion, a sledgehammer rule change was made to crack the nut.

I believe that ACC has concluded that a delay in reporting moves would not help to deter cheating. They have given no reasons, and I find this surprising. If nobody outside the room knows the position then surely cheating would be more difficult. I understand the reasons why this might be commercially disadvantageous, but that's a different question.

[Roger de Coverly]

The ban on writing a move before playing it, which destroyed a generation of training, had nothing to do with Monroi.

It came in at exactly the same time as the Monroi device was legalised and the Monroi would have enabled a player using one to see the position as a diagram immediately after the proposed move. That would have been a good reason to reject the Monroi, but someone, perhaps FIDE got some fee for approving it. The same applied to DGT Clocks ten years earlier, some of the less desirable features of these clocks had to be incorporated into the Laws of Chess because they were the "official" clock and were paying money for the privilege.

[Roger de Coverly]

I believe that ACC has concluded that a delay in reporting moves would not help to deter cheating. They have given no reasons, and I find this surprising. If nobody outside the room knows the position then surely cheating would be more difficult. I understand the reasons why this might be commercially disadvantageous, but that's a different question.

It's a fairly obvious way of non-premeditated cheating that a device just outside the playing area could be used to follow and analyse the game and the suggested ideas communicated back to the player. I know it's suggested that devices are banned from spectator areas as well, but I don't think that's possible when the spectator area is part of a hotel complex to which non-chess guests have access.

[Richard Bates]

The same applied to DGT Clocks ten years earlier, some of the less desirable features of these clocks had to be incorporated into the Laws of Chess because they were the "official" clock and were paying money for the privilege.

It is difficult to imagine which particular features you might be referring to...

[Roger de Coverly]

This might need a new thread, but what is the relationship between the proposed ACC and FIDE Arena? On the website of the latter, I read at
http://home.premiumchess.net/fide-arena ... tages.aspx

AceGuard is a chess cheater’s worst nightmare. It constantly monitors and checks for every possible way a player might seek to gain an unfair advantage – everything from manipulating lag to using a chess engine. The program creates a sophisticated profile of every individual player and quickly zooms in on any suspicious event. Monitoring is immediately increased for any suspect player and a team of anti-cheating experts are called in to examine the evidence before any action is taken. AceGuard's revolutionary technology is supported by special FIDE anti-cheating procedures so as to supply the safest environment possible. It is important to note that action is not usually taken immediately for a single case of cheating. Clear and compelling evidence is collated and reviewed before any decision is made.

All Premium Chess full members and FIDE online arena members are given a fairness index star-rating based on an analysis of all the games they have played, so that at a glance you can see the reliability of your opponent. The fairness index rating is one of the challenge filter options.

AceGuard has a proven track record of success and has been rigourously tested and approved by government authorities, FIDE, national federations and master strength players,

Presumably play on FIDE Arena is deemed to be governed by the over the board Laws of Chess. These don't permit external sources of information such as books, databases and tablebases to be used during play. Unless done covertly, it is not possible in OTB chess to consult external sources whilst in the presence of the opponent, arbiters and spectators. Whilst FIDE Arena probably has built in spyware to detect what other programs are running, it cannot know anything about what's running on unconnected devices or even what traditional paper based books are open. You can use monitoring to note that a player's style closely matches known book moves or previous games, but that's a false allegation if it's just memory, knowledge or understanding.

I'm assuming the list of "approvals" is just marketing hype.

[MartinCarpenter]

I'm a bit sceptical about being able to use a hidden camera and program a computer to parse the photograph. Maybe if you could get an overhead view from something like a demo board. In clothing you'll be getting views from lots of subtly different, often not very clear, angles. Can't imagine not getting a reasonable quantity of mistakes.