Niemann - Carlsen

[David Ottosen]

Every titled player that has ascended from master to Grandmaster has the same decreasing average centipawn score as described above. Now, when we take a look at Hans Niemann's game and analyze all the games he played from when he was 2300 up to his ascension of 2700 the average centipawn score never changes. He has a unique pattern compared to all other ascending Gm rated players in the world.

Just want to be clear if I understand; basically he goes from having a normal 2300 level centipawn average score to suddenly having a 2700 level centipawn score basically overnight, and maintains it for the years since it started? As opposed to the more normal 2300 level score, then slowly getting better?

I ask because my understanding of the first game with Carlsen was that his game score wasn't actually that spectacular and his win was more Carlsen playing well below his usual level. So the question is more "did he jump from 2300 level scores to 2700 level scores overnight and saw the associated rise" or "did he jump from 2300 to 2500/2600 and for some reason, his style of play induces more bad play by his opponents".

[Brad Thomson]

I doubt if we will ever know what technology was used or how it works,

Then you have not demonstrated opportunity and cannot convict.

[Patrick Kirby]

I can't make heads or tails of this Chessbase article https://en.chessbase.com/post/statis...f-hans-niemann. How does playing at a 2500 level, even if true, support allegations of cheating? It seems more to support the hypothesis that David mentioned that his playing style induces opponents to play below their usual level. You can see elements of this in the game Niemann - Firouzja from St. Louis, where Hans blitzed out a crazy looking piece sacrifice and Firouzja responded poorly (but then Hans failed to convert his advantage from a position where any of the other players in the tournament would have been pretty much a lock to get the full point).

Or is the claim that he's only cheating on a few moves a game and the "super-moves" blend in with the statistical noise? Would this even be feasible as a way of cheating? Could you go from 2500 strength to 2700 just by using a computer for 3 moves a game if you were really good at identifying the critical moments? And if you were doing this wouldn't it still show up as a signal in the statistical analysis (presumably your cheating accomplice would only feed you moves where there's a large difference between the computer's top line and the second-best move, so those few moves would end up having an oversized effect on the average and the variance).

To my mind there's nowhere near enough proof here to make a definitive claim of cheating. The proof will come when we see whether Niemann maintains his high level of play in tournaments with robust anti-cheating measures.

[Kenneth Regan]

I did not realize Sid Belzberg would echo my comments on Facebook Messager---a private medium---on this forum. I did say I would check into various Internet stuff today, and in fact Nate Solon today has presented most of what I was going to say: https://zwischenzug.substack.com/p/d...-niemann-cheat See also my comment just placed there.

[Sid Belzberg]

Just want to be clear if I understand; basically he goes from having a normal 2300 level centipawn average score to suddenly having a 2700 level centipawn score basically overnight, and maintains it for the years since it started? As opposed to the more normal 2300 level score, then slowly getting better?

I ask because my understanding of the first game with Carlsen was that his game score wasn't actually that spectacular and his win was more Carlsen playing well below his usual level. So the question is more "did he jump from 2300 level scores to 2700 level scores overnight and saw the associated rise" or "did he jump from 2300 to 2500/2600 and for some reason, his style of play induces more bad play by his opponents".

Every player shows a linear correlation between their rating and their centipawn loss score as well as the overall consistency of each move, the higher the rating the lower the centipawn score and the lower the standard deviation. So far this analysis has held true for every player that has ascended from 2500-2700. Hans Neiman is the only player that shows no improvement in the overall consistency of his moves (ie tends to choose the best move the machine offers or very close to what the machine offers as second best) , this is the centipawn loss that for every other player during their 2500-2700 ascent gets lower and lower but not with Hans Neiman,
This would indicate perhaps that in critical positions he chooses the best move but most of the time has lower quality moves using centipawn loss as a barometer. Given that this pattern is unique does not look good. Apparently, chess.com will have a big announcement this week so we will see what happens.

As one poster put it,

"These correlations for all the players other than Hans is insanely high, so he sticks out like a sore thumb and the likelihood of randomly playing one good move in a game and it not be cheating seems... Small."

[Sid Belzberg]

I did not realize Sid Belzberg would echo my comments on Facebook Messager---a private medium---on this forum. I did say I would check into various Internet stuff today, and in fact Nate Solon today has presented most of what I was going to say: https://zwischenzug.substack.com/p/d...-niemann-cheat See also my comment just placed there.

We discussed centipawn loss analysis that, understandably, you implied you were too busy to look at at the moment but would take a closer look at it when you had time. The articles you mention are not related to what we discussed.

[Patrick Kirby]

Every player shows a linear correlation between their rating and their centipawn loss score as well as the overall consistency of each move, the higher the rating the lower the centipawn score and the lower the standard deviation. So far this analysis has held true for every player that has ascended from 2500-2700. Hans Neiman is the only player that shows no improvement in the overall consistency of his moves (ie tends to choose the best move the machine offers or very close to what the machine offers as second best) , this is the centipawn loss that for every other player during their 2500-2700 ascent gets lower and lower but not with Hans Neiman,
This would indicate perhaps that in critical positions he chooses the best move but most of the time has lower quality moves using centipawn loss as a barometer. Given that this pattern is unique does not look good. Apparently, chess.com will have a big announcement this week so we will see what happens.

But as I mentioned in my last post, wouldn't systematically cheating in critical positions still create a signal in the data?

Assuming that "critical" positions are ones:
1. Where human players are more likely to make mistakes; and
2. Where the difference between the best move and the second best move is significant;

It seems like cheating in these positions would still significantly increase average move quality compared to non-cheating players, even if it was only a few moves a game.

[Kenneth Regan]

Sid, those articles I cited---my own---are absolutely relevant as put in my replies to Nate Solon. I categorically disavow any support that might be implied for the sources you queried me about on FB Messenger.

[Sid Belzberg]

Sid, those articles I cited---my own---are absolutely relevant as put in my replies to Nate Solon. I categorically disavow any support that might be implied for the sources you queried me about on FB Messenger.

I apologize, Ken, I only read the article and not the comments you made that clearly referred to centipawn loss analysis. As mentioned originally when you showed me this article, Hans Neiman's Graph of his rating versus centipawn loss during his ascent (in particular when he ascended from 2500-2700) is entirely different than all other grandmasters that achieved these lofty heights..
Your thoughts on this are what I was seeking, as I mentioned in this forum and on your messenger, and still am. As I mentioned, the code and methodologies, and data sources are entirely disclosed and reproducible.

I can assure you that I am not alone in the chess world or the data science world in this request I posed.

https://en.chessbase.com/post/statis...f-hans-niemann

[Tom O'Donnell]

Suppose you are playing a game against a known cheater. Your opponent makes what looks like a double-edged sacrifice. Would the thought not cross your mind "I wonder if this guy is using an engine right now?". I think it would for many people even if there is no evidence their opponent is using an engine.

One possibility is that Niemann is using this uncertainty to his benefit and is strategically gambling more over the board.

It could also be that his opponents are genuinely unsure about what is going on and are simply playing worse as a result. Their attention is divided between playing good chess and wondering if they are being cheated at that moment. Even if there is zero evidence they are being cheated.

I have to say I don't buy the "we need evidence beyond a reasonable doubt" standard. Civil cases for millions of dollars are determined by preponderance of evidence. Maybe split the difference and say if it's 75+% someone cheated, they get the hook. I guess the big question is what exactly constitutes evidence, vs conjecture/confirmation bias.

[David Ottosen]

Every player shows a linear correlation between their rating and their centipawn loss score as well as the overall consistency of each move, the higher the rating the lower the centipawn score and the lower the standard deviation. So far this analysis has held true for every player that has ascended from 2500-2700. Hans Neiman is the only player that shows no improvement in the overall consistency of his moves (ie tends to choose the best move the machine offers or very close to what the machine offers as second best) , this is the centipawn loss that for every other player during their 2500-2700 ascent gets lower and lower but not with Hans Neiman,
This would indicate perhaps that in critical positions he chooses the best move but most of the time has lower quality moves using centipawn loss as a barometer. Given that this pattern is unique does not look good. Apparently, chess.com will have a big announcement this week so we will see what happens.

I guess what I mean is this; say a 2300 has an average centipawn loss of 40, while a 2700 has an average centipawn loss of 10. In the normal progression you describe, you see something like a player go from 40 to 35 to 30 to 20 to 15 to 10 (or whatever) as they progress from 2300 to 2700. My question is whether Niemann was a normal 2300, happily losing 40 centipawn per game, and then overnight switched to losing 10 (bypassing the normal progression) or whether he suddenly improved to losing 25 per game, but saw an improvement in his results that seems unusual for this level of play; ie, does his centipawn loss rate indicate that he's a real 2700, or does his centipawn loss rate indicate he's a 2500 who through a statistically improbable run of bad play by his opponents has reached 2700?

[Peter McKillop]

.....
I have to say I don't buy the "we need evidence beyond a reasonable doubt" standard. ......

Here's another log for the fire:

https://en.chessbase.com/post/wall-s...likely-cheated

[Sheldon Ab]

After watching the video on the computer analysis I'm starting to think he was cheating. The evidence seems convincing but the big question for me anyways is how did he cheat? I'm only asking in regards to the live games. That's what I can't wrap my head around.

[Sid Belzberg]

I guess what I mean is this; say a 2300 has an average centipawn loss of 40, while a 2700 has an average centipawn loss of 10. In the normal progression you describe, you see something like a player go from 40 to 35 to 30 to 20 to 15 to 10 (or whatever) as they progress from 2300 to 2700. My question is whether Niemann was a normal 2300, happily losing 40 centipawn per game, and then overnight switched to losing 10 (bypassing the normal progression) or whether he suddenly improved to losing 25 per game, but saw an improvement in his results that seems unusual for this level of play; ie, does his centipawn loss rate indicate that he's a real 2700, or does his centipawn loss rate indicate he's a 2500 who through a statistically improbable run of bad play by his opponents has reached 2700?

No, the interpretation of the result shows that Nieman's average centipawn loss and standard deviation during his ascent from 2500-2700 is entirely different from other GMs with similar rating increases during the 2500-2700 period. Because the general quality of his moves was lower than other GMs during their ascension, it could be interpreted that he received assistance at critical junctures hence the general quality of his moves was comparatively poor compared to his peers, but the odd critically important move pushed him over the top.

[Pargat Perrer]

.... hence the general quality of his moves was comparatively poor compared to his peers, but the odd critically important move pushed him over the top.

Maybe the top rated players on this board can answer this question: Is what Sid just described a viable way to win consistently and get to 2700 (assuming your cheating method goes undiscovered and uninvestigated)?

I have to think it would NOT be viable ... because for example you are playing let's say 4 consecutive moves that are slightly inferior to those of your super-GM opponent's 4 moves and then on the 5th move you use your cheating device and find the absolutely best move at the critical juncture .... but you are playing this move from an INFERIOR POSITION, are you not? That means you are constantly having to come back from inferior positions, game after game, time after time.

Can that work to get one to a 2700 rating? Again, assuming the cheating method works and you correctly assess when you should use it, i.e. you correctly determine the "critical junctures".

EDIT: if what Sid proposes is indeed what has been going on with Neimann, it should show up .... do a move-by-move score assessment of each of his games and graph it .... you should see that he falls behind in every game, and then "miraculously" makes a single move to either get back to equality or even get ahead. This should be EASY to detect, and it should be happening like a repeating and regular cycle in EVERY ONE of his games.