Carlsen resigns on move 2

[Paul Cooksey]

I find the chessbase article unhelpful too. I think you need to a correlation between unusual data and computer play before it becomes interesting to analyse in this context.

I'm plodding along trying to understand the statistical model so I can respond to some of Roger and Nick's points. It seems to me extremely unlikely a 2800 player has had computer assistance in over the board play. But if I understand the Regan methodology I will never be more than 93% sure he cheated, because the best computer is only giving me a 3 sigma benchmark. If I have a lot of data I can be very sure it is a 93% result, but no amount of data takes it higher. That is important to me, because 93% is below my threshold for reasonable doubt.

Maybe as the statisticians have said, algorithms are most useful if identifying suspicious results rather than in providing evidence. But reinforces my view that physical security at elite events is very important. Assuming I reject Justin's position of course, which I am more than happy to do.

I realise in separating an assessment of "if this data is good, this is the statistical likelihood that cheating is happening" from "this data set is good enough to give a reliable result" I'm in danger of looking like I am saying "Sex Panther, sixty percent of the time it works every time". But I think Roger and Nick are objecting in different ways.

Nick seems to be disputing the assumption that unexpected performance correlates to a likelihood of cheating. I don't have the stats to argue on this point. But I think it is taking on Regan directly in his field of expertise, so as a shill for authority figures I would need a lot of convincing.

Roger's objection seems to be to whether the data is ever reliable enough to allow an algorithm to give reliable results. I'm thinking about this because I think it is the area where our collective expertise in chess is relevant. Clearly Regan understands the objection, because he went to some length to say how proud he was about being able to predict the improvement of juniors, even when covid prevented them playing. But it does seem an important point to me.

[Geoff Chandler]

What method of signalling do you propose

Today at a top level event very difficult but in the past before these stricter measures
came in, very easy. Even the yoghurt in '78 could have been used, though that was too blatant
(which is why is could be perfect.)

With the security as it and I wanted to get a signal to a player today playing in a WC final in a glass sound
proof cage and remember I only have to do this once to tell the player a good move is available.
(though the idea I have in mind could be used twice.)

Yes! (first I need to speak to someone regarding the layout of the playing booth.) but it's perfectly feasible.

[JustinHorton]

and remember I only have to do this once

No you don't, you're trying to put 100 on his Elo

it's perfectly feasible.

And yet we are no closer to discovering how

[MartinCarpenter]

I find the chessbase article unhelpful too. I think you need to a correlation between unusual data and computer play before it becomes interesting to analyse in this context.

I'm plodding along trying to understand the statistical model so I can respond to some of Roger and Nick's points. It seems to me extremely unlikely a 2800 player has had computer assistance in over the board play. But if I understand the Regan methodology I will never be more than 93% sure he cheated, because the best computer is only giving me a 3 sigma benchmark. If I have a lot of data I can be very sure it is a 93% result, but no amount of data takes it higher. That is important to me, because 93% is below my threshold for reasonable doubt.

Ah, that Elo/sigma isn't remotely fixed. Think about it. A 2800 playing *1* game at 3300 strength isn't at all suspicious. One playing a short tournament at that level probably isn't either. Several tournaments in a row at that level will indicate either a cheat or an unprecedented genius.

The model has to reflect that, and the elo differnce for 1 sigma does that. They'll shrink as you add more games in. I think someone earlier said 1 sigma was +-200 elo for 6 games, it'll be far less for 10, then less again for 20 etc etc.

In general of course there's a fair chance that your initial estimate of the players strength was just nonsense. That can happen, if not at 3300 Such is life I'm afraid. Very useful at a population level whatever, cf: https://www.4ncl.co.uk/data/Article_on_Fair_Play.pdf

[JustinHorton]

if what Chessdotcom allege is correct, and Hans cheated online more than he has admitted, do you think this was (a) before he was banned at age sixteen or (b) after, and what would be the implications of either?)

No takers to this, but it's bugging me, so let me share my reasons.

My working assumption is that the answer is not (b), unless by amazing coincidence Chessdotcom decided Hans was cheating on their site in the early days of September this year. But unless that's the case, he would surely have been banned already, and he wasn't. Does that make sense?

But if the answer is (a) and all Chessdotcom are saying is that Hans had been cheating, when or before he was sixteen, a lot more than he has admitted in public ("I wanted to gain some rating, I just wanted to get higher-rated so I could play stronger players, so I cheated in random games on Chesscom") then I'm not seeing relevance. I'm very prepared to believe it, in fact I'm prepared to assume it, but so what? They allowed him access to his account again, they were (as he observed) very happy to welcome him to their "Global Championship", so what's it go to do with anything? He can't be welcome in August, and not welcome in September, unless you either do something bad between those two dates, or something bad about you is discovered between those two dates.

But if it's just that he's been fibbing about the extent of his cheating in 2019 - the extent of his cheating didn't matter to them in August 2022, so why did it matter in September?

[JustinHorton]

(I'm probably saying nothing new above, but I'm just trying to tease out how weird it is, on the face of it, for Chessdotcom to have made that statement, and perhaps for them not to have been asked more closely what they meant by it. I know one answer to my question is that we may know more when they release whatever information they're going to release, but at the same time, it might be worth observing now that if Hans Niemann cheated in a thousand games three years ago, this would raise as many questions for Chessdotcom as it would for him.)

[Paul Cooksey]

Ah, that Elo/sigma isn't remotely fixed. Think about it. A 2800 playing *1* game at 3300 strength isn't at all suspicious. One playing a short tournament at that level probably isn't either. Several tournaments in a row at that level will indicate either a cheat or an unprecedented genius.

Thanks. I still haven't completely understood it, but appreciate the help. I am genuine interest inn understanding the stats better, so book recommendations welcome.

A separate point, a 2800 can't play several tournaments in a row at 3300 and stay 2800 which seems a complication, because clearly you have to catch people before they convince you their baseline is higher than is true.

[Paul Cooksey]

More excitingly than my maths struggles, chess.com data on Hans released through the WSJ.

Many more online games, but still before he turned 18 seems the headline

[Alistair Campbell]

I think Justin’s questions are interesting – it’s a shame I can’t just flick to the back of the book to look up the answers . – but I’ll have a go.

The actual answers may depend on the motivation to cheat in the first place. There could be several reasons, including:

• Reasons of (Self) esteem – look at me – I’m great!
• Other Rewards – (small) cash prizes
• Taking a shortcut – I could have studied a bit harder (and I will in future) but I’ll just look up the right answer now
• Equalising opportunities/gaining access to an exclusive arena – if only I could get to the next level I could stay there, or even thrive
• Getting what you deserve – I’ve played well, I don’t deserve to throw it all away with a blunder. Or even, I’ve put all this effort in, I deserve to be rewarded.

A GM may have different motivations from a patzer.

Q1 – net effect of having an alert system.

I think Jon Tait suggested a figure of +200 over a couple of seasons – but this is more having a blunder check, which isn’t what we are talking about.

My a priori guess might be a half-point every 5 games, which is 5 Elo points? So 100 points in a 100 game season? Does that seem about right? Dull positions may be less likely to have “key moments”, and a GM might spot most of them anyway. Of course, if it’s one full point every 5 games, that’s twice as much(!)

Q2 Probability of being caught. Will be clearly dependent on method, and detection. Detection seems to be based on suspicious behaviour and suspicious results. The “smart theory” of cheating suggests that detection can be evaded by throwing in some sub-standard moves which would be the case if one has already chucked an advantage. Having a second crack at the whip might be relatively immaterial.

Q3 – the “no buzz” situation. The extra information should be helpful – if you think Nxf7 wins on the spot, the fact that there is no buzz may again lead to deeper calculation. Again practice would help.

Q4 (a) – do you cheat forever? I think this depends on the motivation. In theory, I think the plan would be that once the benefits are achieved, then there is no longer any need and use can be reduced to zero. In practice, temptation would be too great.

Q 4 (b) – negative effect of worry. Again, it may be dependent on motivation. If your motivation is to “cock a snook” at the establishment, that may persist and you may get a kick out of continuing,

I feel I’m just scratching the surface here. I suspect that cheating may be the sort of thing that is difficult for GM Grifter to stop without giving up the game altogether; and that the notional gains (see the answer to question 1) aren’t really worth it. Of course, if you were to recalibrate (and the temptation must be to do so) the definition of a key moment to +1 (say) the figures may prove to be much more persuasive.

There should be parallels with cheating in other sports such as golf. And has anyone ever seen “Quiz Show” – stars John Turturro who was in “The Luzhin Defence” and “The Sicilian”…?

[Roger de Coverly]

He can't be welcome in August, and not welcome in September, unless you either do something bad between those two dates, or something bad about you is discovered between those two dates.

They made an offer for the play magnus group in August 2022 which could possibly be relevant.
https://chess24.com/en/read/news/play-m ... -com-offer



Another idea is that they took a more intense look at the history of games on their servers, given that he was becoming a higher profile player. Perhaps they found that suspicion of external assistance wasn't limited to games where the only reward was rating points.

[JustinHorton]

Talking of coincidences, here's the WSJ, annoyingly paywalled, with details of Chessdotcom's investigation.

[Roger de Coverly]

The actual answers may depend on the motivation to cheat in the first place.

At a much lower level, what were we seeing in some of the online 4NCL tournaments? You had veteran players making what appeared a lifetime best performance, only to have their results anulled. Other than future opponents and organisers remembering what happened, there was no obvious adverse publicity.

[MartinCarpenter]

Ah, that Elo/sigma isn't remotely fixed. Think about it. A 2800 playing *1* game at 3300 strength isn't at all suspicious. One playing a short tournament at that level probably isn't either. Several tournaments in a row at that level will indicate either a cheat or an unprecedented genius.

Thanks. I still haven't completely understood it, but appreciate the help. I am genuine interest inn understanding the stats better, so book recommendations welcome.

A separate point, a 2800 can't play several tournaments in a row at 3300 and stay 2800 which seems a complication, because clearly you have to catch people before they convince you their baseline is higher than is true.

It very definitely is an issue that, yes At least in the hypothetical case of someone playing consistently at 3300 you can be pretty sure they're not doing so honestly!

With people at lower grade bands, especially juniors etc, it must be a bit of a nightmare to be honest. Just have to be really cautious I suppose.

[Matt Bridgeman]

Talking of coincidences, here's the WSJ, annoyingly paywalled, with details of Chessdotcom's investigation.

I think it states he likely cheated in 100 online games (25 of which he streamed), including 4 titled Tuesdays, a Pro League (12 of 32 matches there) and into his 17th year, so contradicting Han’s statements on age and that they were just ‘random games.’

In regards to OTB an extract;

‘The 72-page report also flagged what it described as irregularities in Niemann’s rise through the elite ranks of competitive, in-person chess. It highlights “many remarkable signals and unusual patterns in Hans’ path as a player.”

While it says Niemann’s improvement has been “statistically extraordinary.” Chess.com noted that it hasn’t historically been involved with cheat detection for classical over-the-board chess, and it stopped short of any conclusive statements about whether he has cheated in person. Still, it pointed to several of Niemann’s strongest events, which it believes “merit further investigation based on the data.”’

[Roger de Coverly]

Talking of coincidences, here's the WSJ, annoyingly paywalled, with details of Chessdotcom's investigation.

It mentions suspicions about his OTB Elo rise. Whatever it was, surely it was over an abnormal period of lockdown when OTB tournaments weren't taking place?

This is the history showing monthly gains or losses
https://ratings.fide.com/profile/2093596/calculations

He's played a lot of tournaments. I'm not sure that the longstanding exploit of the Elo method still works with monthly calculations. It used to be the case and still is that if your rating was lower than your strength, you gained points to bring the two in line. If you played a lot of games it was possible to overshoot and get a higher rating that strictly deserved. That was a method employed by some Norm seekers back in the days of half yearly or quarterly ratings.