Shocking Flaw in the Hans Niemann ‘100%’ Game Analysis

This is a genuinely shocking Flaw in the Hans Niemann ‘100%’ Game Analysis which Yosha put out earlier this week, supported by Gambit-man. Very keen for her thoughts on this as well as Hikaru’s because this chess game analysis completely confuses me.

💲If you would like to support the channel with a donation then please use: paypal.me/EpicChess

or consider joining the channel to gain access to some awesome perks:

As always thank you for your support!

#hansniemann #niemann #chess #chesscheating #epicchess

331 Comments

  1. now let see if Hikaru will change his tune..again..for the sake of contents.

  2. As long as the same methodology is applied to lots of players differences should be helpful in discovering cheaters. You seem a little hung up on the 100% mark when in fact anything over 90 is actually ridiculously rare. Stop dwelling on exactly 100%. The overall trend analysis discoveries don't have to get hung up on the 100% figure. I think you may be overreacting to that fact, step back and take a look at the big picture. This like you suggested is the start of a more sophisticated big data analytical review not the end. Thank you for participating in the analysis.

  3. It doesn't strike you as unusual that using the same tool and methodology, that Hans has many many more 100% "accurate" games then all other GMs though? Even if what you're saying is true it is still strange that he is such a big outlier

  4. It was Hans rambling interview that really sealed the fact he is a filthy cheater. Saying this is the truth 100 times. Saying he deleted all social media to reference social media later. Saying things like “ completely human move”. Who does that? How gullible do you have to be to think that guy is clean?? Smh

  5. All u did is nothing no analysis just adding confusion

  6. Fair criticism, but at the end of the day I don't know if it's really important. Even if the analysis method isn't perfect, if you find a statistical difference between Hans' games and other GMs it's a red flag. If you analyze games of players that should be in the same ballpark, using the same method (even if not perfect), you should get similar results. If the results don't match either Hans is on a whole different level, or he's cheating.

  7. you gotta understand that those engine are better than any human brain so no matter which one or how low the depth is if one of them say its a good move then its a good move it doesnt have to be stockfish 15 46 depth top engine move to be considered a good move, stockfish 15 depth 36 is still better than every human.
    btw depth 22 is already better than every human.
    so the Analysis is still valid.

  8. Run Stockfish 11, pawn to h6 shows up as the 1st move for me at depths 21, 22, 23 and 1st or 2nd at multiple other depths around there. It's 3rd best at higher depths.

  9. This is why intellectuals are bad. Some intellectuals have data to show he is cheating and some intellectuals have data to show he is not cheating. Let’s go back to how things are done in the good old days which is using common sense. Magnus thinks he is cheating and I would say most GM thinks he is cheating. He cheated before. He is a friend of a known cheater. To me, he cheated. But in the court of law, he is innocent until proven guilty and as far as I know, nobody can definitely prove he is cheating.

  10. what stops Hans from mixing outputs from multiple engines to not make it COMPLETELY OBVIOUS that he is cheating? I can't be the only one thinking this way…

  11. That was a terrible and superficial game played by white. Bxh6? followed by Bxg7??, which are not shown in this video, are terrible blunders by white. From then, Hans won with simple and forcing moves that could have been found even by a 2000 ELO rated player.
    Hans was not cheating in this game in my opinion. There are other games where Hans' play is suspicious

  12. Pretty sad that people are agreeing with this yosha without actually looking at the evidence themselves or trying to understand what it means…

    Just standard internet things.

    Thanks for actually having a brain!

  13. "I don’t know if Niemann cheated OTB or not.

    The evidence I’ve seen so far is far from proof and, in most cases, is junk. Look at all the "experts" who have embarrassed themselves with the ChessBase "Let’s check" feature, for instance."

    "MY MOTIVATION HERE IS THAT THERE ARE A TON OF PEOPLE ABUSING "DATA" AND "STATISTICS" TO FIND THE CONCLUSION THEY WANT.

    AND THEY DO IT WITH TINY DATA AND CHERRY PICKING."

    "80% COULD RISE TO 90% AND THEN 100% IN A MATTER OF DAYS.

    THE GREATER THE NUMBER OF OBSERVERS THE HIGHER THE MAGIC GUILT NUMBER BECOMES."

    "Because of all the detectives who have gone through Hans' games there are literally more than 100 engines used in the analysis so far.

    THIS MAKES THE 100% SCORES COMPLETELY IRRELEVANT !"

    Live demonstration of the above:

    Why Engine Correlation doesn't work – The "most incriminating" evidence against Hans Niemann refuted

    https://www.youtube.com/watch?v=GV28HO2Ea_s

    So those with a nut-like mind – who immediately believe everything they hear or see, without thinking – should be silent.

  14. There is still a major problem when only Hans has games that correlates so much relative to all other players.

  15. Even if that isn’t a proper analysis, the fact that no other player has so many games with 100% correlation may be indicative of something.

  16. It's funny how you say you don't understand but still come to conclusions

  17. Yup, seems like a bit of a cluster fk this mess… It's easy to jump on the hate bandwagon but as you said, it has to be done right… Witch hunts are no fun…

    On the other hand, you have to realize, if some one is cheating, say Hans Niemann, it has to be done with time limits on whatever engine(s) he's using and he can freely reject those suggested moves anyway at any point for whatever reason.

    I heard there is an official investigation starting, which after all this background witch hunt seems like a god send, because this is not just a video game cheater and the stakes from false accusations could be high.

  18. The problem is that it does not prove what it tries to prove.
    Like let's say he got those numbers of 100%. That means he probably uses an engine for all his moves.
    But then when it is brought up that he would have much better results than this, people then switch the goal post and say he would only need to use an engine some of the time, on a few moves per game, which probably would not be enough to get him to 100% and if it does, then he was already at 90% or more.
    People have to choose a narrative and stick to it, because right now those theories are just impossible taken all together. People are simply fishing to prove their own bias. Maybe he is just that good and is better than Carlsen. Maybe he played a lot with engines and play like them unlike older people.

    I don't know, the whole thing seems silly nonsense. It's quite likely a lot of people are cheating online and he is not really good at hiding it. The real question is about the over the board games. Are all those 100% games over the board? if that it the case, unless he cheats on all the moves, it's pretty hard to not conclude he is pretty good or at least engine level good. In other words, those analysis may just prove he is the best. 😁

  19. how often do other people get 100% is the engine consistent? has magnus or hikaru ever gotten 100% on that engine? Don't they play thousands of games on the site too? Surly GMs on those caliber can also get these errors once in a while. I do remember seeing a recent video from hikaru saying he has never gotten close to 100% in any of his games.

  20. If it is consistently applied across a sample of GMs and SuperGMs, the data should look consistent across players. You could prove this data method to be flawed by applying the methodology to Magnus' and Hikaru's games to generate an unusual amount of 100% perfect games or 90+ % games. If after doing that, Hans is still a notable outlier, then it's likely that Niemann is cheating.

    There's no reason to believe that a random sample of engines would specifically target Niemann more than any other Gm/SuperGM. The reason this method has validity is because of the ability to check variable depths. If you use one depth, a cheater only has to fall beneath or exceed that depth to look like a normal player but by having multiple depths a cheater would have a more difficult time not being somewhere in the range of AI.

    If the method is flawed and legitimately problematic then any GM/SuperGM could be caught by it. If Niemann is the only one being caught by it, then it's still likely a valid method.

  21. It’s not a flaw if this is applied to every chess player equally, it would give false 100% accuracy to other players as well but hans still remains an outlier

  22. Why do you think Hans had so many more 100% games compared to other GM's? Would a faulty analysis applied to all players then become invalid anyway, or would it still be a valid comparison? I'm not very clued up on statistics.

  23. I'm not much of an expert I only have an undergraduate math degree with 2 years of mathematical statistics courses. So I can't say much about in detail but I can say that EVERYTHING in that video just appeared to me to be outright wrong. From the methodology, the probability calculations, the comparisons being made, the data being used. Every single thing she said my intuitive reaction was just like, huh?? That looks… wrong.

  24. Fabi also stated multiple times that if you look from a statistical or mathematical point of view, we should not credit Yosha's analysis as valid because she is neither a data analyst nor a mathematician. Which I completely agree with

  25. Niemann won the game in 22 moves so maybe engines see now h6 as a great human move, hence the 100% 🙂 By the way Dlugi claimed that it would be only necessary for a cheater to use help from the outside 3 or 4 times during a game so if you want to cheat cleverly you don't use it on every move as a lot of human moves are pretty automatic

  26. I think the original video made it clear how the check took worked, about how it looks at multiple engines and any match will show positive. But applying this same tool to other top players, you don’t see as many 100% games, and you also see a statistically significant lower average. That’s what I think the main point is. That Hans average is an outlier.

  27. "If it weighs as much as a duck, it must be wood and therefore a witch!"
    "Huh?"
    "Well it may not be proof but it is evidence."
    At what point do we throw out bad science? Why is everyone so quick to dismiss Regan but accept this armchair analysis with methodological flaws?

  28. It's a comparative analysis though… not an absolute one. 100% is the score on the scale defined by the depth of the engines used. It's still a fair comparison when you look at how other very high rated GMs simply scored lower, on the same scale, when as much shouldn't necessarily be expected based on rating, experience, games played etc. I for one, do not believe Hans is a super genius, and certainly not about to become a future chess world champion. I would be thoroughly impressed if he ever manages to do so, don't get me wrong. But I just don't think we should expect him to wipe the floor with some of the best chess players alive as if it is nothing.
    By the way, the fact this 100% can still 'be accurate', also has to do with how for any position on the board, there really is no definitive 'best line' to play. This is due to how 'perfect chess', is actually a draw (or at least very draw-ish, because the rules are fair and the first move advantage is statistically negligible). The deeper the engine calculates, the higher the potential for a new pathway to yet another draw will be found. We therefore should not even expect a 100% accurate 'winning' line. There should be countless of variations all leading to a (probable) draw. A strong engine correlation might definitely reveal a suspicious game. But it's no direct evidence of cheating. I do have to say though to claim it is 'statistically irrelevant' is really just not understanding what it does mean. That's like arguing a total beginner at chess could accidentally stumble across all the best moves according to engines and therefore it wouldn't be suspicious. That's really not what it means. This is not about one single move per game. This is about engine correlation averages over lots of moves and entire tournaments.

  29. I ran another “100%” game from Hans on stockfish and there were multiple inaccuracies. I know this means probably some other engine liked the move but it was still like you said -0.5 to -0.7 swings. To say that’s 100% accuracy is, well, inaccurate, or at least there’s a massive caveat.

    My only question is why were other players’ games way lower %. Did they run fewer engines against their moves?

  30. Just to let the crowd know, the 'depth' is usually set to a high number, perhaps 1 million, and let stockfish exhaust the tree. Interesting magic numbers you're using, 46, 44, etc. Let it run overnight, give stockfish some lung capacity and let it explore.

  31. . . . represents King or Rank1 or File A
    . . .. represents Queen or Rank2 or File B
    . .. . represents Bishop or Rank3 or file C and so on.

    Piece, file, rank is the order of 3 sets of signals .. . .. Pawn .. . . e . .. .. 4. If Niemann is receiving 9 bits of information and interpreting in that order through any means he can be given the best move at any point.

    Place him in a Faraday cage and see how he plays!

  32. ChessBase clearly states do NOT use the "Let's check" feature to detect cheater. The number of engines used and results will vary widely for each game, so it is impossible and useless to compare percentages between players. If someone uploads analysis from a 20-year-old engine that picks the h6 move in this video, then it will count towards the 100%, even though modern engines say it's a bad move. Useless for cheat detection!

  33. They used the SAME methodology to calculate the statistical percentages. They didn't used this method to make Hans look bad or cheater. Even though there might be calculation mistakes etc this algorithm shows good enough that none other than Hans has achieved above 90-100% games for 10 games. Could you come up with a custom set of calculations that makes Fabi look like an above the average? During those time period Hans was between 2400-2600 so even 2750+ rated players couldn't achieve that kind of a score according to this system.

  34. The ChessBase "Let's check" method is not consistent between players! For example, 100 different engines and depths could have been applied to a Hans game, while 10 different engines and depths applied to a Gukesh game. This gives a MUCH bigger chance that each of Hans' moves will match at least one of the engines. Comparing the number of 100% games between players is USELESS. ChessBase itself tells us this, don't use this method to catch cheaters!

  35. great some people are thinking about the statistics hear

  36. Hikaru is juicing this Hans Drama and has a lot of bias against him from the start. Many other youtubers are also exploiting the same thing.

  37. I'm under the impression that this 100%-stuff are rumours. I follow Agadmator and a few others and I'm pretty sure I would have seen a video on it immediately, had it been proven.

  38. Shocking flaw in your analysis is that you use S15. Of course Hans does not play S15 moves in 2020 even if he cheats… common man! At least use that SF version that would potentially be used to cheat in 2020.

  39. Fair point and would be realy usefull moving forward for these things to be clarified, alas this doesnt change the fact that Hans' stats with this (flawed?) analysis are still better than any chess player ever. As far as i understood from Yosha's video, the same exact (probably flawed) method that was used to evaluate Hans' games was used to extract the stats from all the other players on that video. So maybe Hans' stats were bumped up due to the flaw in method, but so was everyone that he was compaired to… Plus a more recent video (watched Hikaru's reaction to it but dont remember the name of the vid) showing Hans' lack of actual improvement as his elo got higher compaired to every other superGM is pretty damning for Hans… Essentialy the video shows that Hans' playing profile fits to a 2500 player while his results have him at 2700, and also shows that that is not the case for every other 2700+ player. Im just w8ing for the reveal of HOW he cheated OTB at this point. If i didnt mention it, im 99% sure that the video is about OTB games and not online, if im not mistaken.

  40. Not so shocking after all, haha. But nice clickbait

  41. Has anyone applied the same analysis against other renown GM – such as Magnus Carlesen’s game play?

  42. Really want Hans Neiman to “allow” Magnus Carlsen to be open about what would like to say but won’t unless Hans Neiman will “allow”. Clearly legal implications of being sued by Neiman are resulting in Magnus remaining silent on this. Would really like to hear what Magnus would want to say openly.

  43. Apparently openings/known theory do not count in analyzing the game as anyone can learn them. Also games without enough moves past know theory do not count, which makes sense.

  44. Thanks for highlighting this. The analysis that Hikaru and everyone else is citing has absolutely no statistical basis.

  45. lol this is shocking, falsely incriminatig someone with bogus science like that. Good on you for exposing this

  46. so which engine is hans being accused of using?

  47. However flawed the methodology of using the ChessBase tool might be, saying "here is a single move that is inferior to the depth 46 Stockfish recommendation, so this proves he wasn't using Komodo at depth 14" is certainly even more flawed. This is so painful how everyone with zero knowledge about statistics feels the need to give their "expertise" on this topic…

Leave a Reply

Your email address will not be published. Required fields are marked *