Dear Ken, I’ve long been an admirer of your work (and of you personally), and still am. Your confident declarations about the lack of evidence for Hans Niemann’s having cheated against Magnus Carlsen in the Sinquefield Cup or even, as far as you can tell, in any games since 2020, has played an important role in my assuming Niemann’s innocence not just as a legal and moral matter, but as a matter of fact.
1. Yes, a large tournament staff once tried this as a trial, but the results were pell-mell and gave too little data. A few games were lost before turn 9, forgetting I only start at turn 9 anyway.
2. Although the novelty per ChessBase Cloud + Mega was 10.Qxd4, I dropped turns 1-14 until MC's long think at turn 15. Redoing from turn 10 made little difference.
3. No one has been able to make a me a version of Leela that runs *in batch mode* on Linux/UNIX like Stockfish and Komodo/Dragon can. My requests for such compiles of Fritz and Houdini have been turned down. My model is not predicated on the "which engine?" question anyway.
4. The one speculation I have ventured in public (in Chris Chabris's item on Facebook and some interviews) is that insofar as the one unusual element was what Niemann called the "ridiculous miracle" of his having previewed the variation with 13...Be6! shortly before the game (the cheating past having been well supposed), more causative weight should be given to iit. Apart from my speculation, I think Tyler Cowen has the right take at https://marginalrevolution.com/marginalrevolution/2022/09/chessdrama-splat.html As for "feel", I cannot judge---except to say that my model finds the game relatively clear-cut, in accord with what several GMs have opined.
5 and 6 are private. On your last paragraph, part of the reason it applies is that I have deliberately kept things simple. The original name "Fidelity" for my site https://cse.buffalo.edu/~regan/chess/fidelity/, besides being a pun on FIDE and (playing in good) faith and being a synonym for concordance, referenced my original intent to employ distributional distance measures, of which /fidelity/ is one. Indeed, quantum fidelity was the basis of the statistical test by which Google claimed to have achieved "quantum supremacy" in October 2019, as I covered in the article https://rjlipton.wpcomstaging.com/2019/10/27/quantum-supremacy-at-last/ But I quickly switched to a simple "Bernoulli trials" model, abiding a necessary post-hoc adjustment for chess moves not being fully independent, because although maybe not as sharp it is robust and not cranky. I also purposely avoid any use of minimization over the small data of a player's own moves, again trading sharpness for avoiding greater issues of "what are the error bars on your error bars?" Nor do I use sequential information or analyze differences between engines at a level that would fuel a dedicated test---two things I have encouraged online platforms with their greater human and computer resources to do. The point is, these ideas are out there for someone able to bring them into play (plus something I mentioned on the podcast with Altucher is my own venture into sequencing and a form of minimization-on-small-data I have less objection to), so your putative smart cheater would have more to consider.
Greetings, Dennis! Here are brief answers:
1. Yes, a large tournament staff once tried this as a trial, but the results were pell-mell and gave too little data. A few games were lost before turn 9, forgetting I only start at turn 9 anyway.
2. Although the novelty per ChessBase Cloud + Mega was 10.Qxd4, I dropped turns 1-14 until MC's long think at turn 15. Redoing from turn 10 made little difference.
3. No one has been able to make a me a version of Leela that runs *in batch mode* on Linux/UNIX like Stockfish and Komodo/Dragon can. My requests for such compiles of Fritz and Houdini have been turned down. My model is not predicated on the "which engine?" question anyway.
4. The one speculation I have ventured in public (in Chris Chabris's item on Facebook and some interviews) is that insofar as the one unusual element was what Niemann called the "ridiculous miracle" of his having previewed the variation with 13...Be6! shortly before the game (the cheating past having been well supposed), more causative weight should be given to iit. Apart from my speculation, I think Tyler Cowen has the right take at https://marginalrevolution.com/marginalrevolution/2022/09/chessdrama-splat.html As for "feel", I cannot judge---except to say that my model finds the game relatively clear-cut, in accord with what several GMs have opined.
5 and 6 are private. On your last paragraph, part of the reason it applies is that I have deliberately kept things simple. The original name "Fidelity" for my site https://cse.buffalo.edu/~regan/chess/fidelity/, besides being a pun on FIDE and (playing in good) faith and being a synonym for concordance, referenced my original intent to employ distributional distance measures, of which /fidelity/ is one. Indeed, quantum fidelity was the basis of the statistical test by which Google claimed to have achieved "quantum supremacy" in October 2019, as I covered in the article https://rjlipton.wpcomstaging.com/2019/10/27/quantum-supremacy-at-last/ But I quickly switched to a simple "Bernoulli trials" model, abiding a necessary post-hoc adjustment for chess moves not being fully independent, because although maybe not as sharp it is robust and not cranky. I also purposely avoid any use of minimization over the small data of a player's own moves, again trading sharpness for avoiding greater issues of "what are the error bars on your error bars?" Nor do I use sequential information or analyze differences between engines at a level that would fuel a dedicated test---two things I have encouraged online platforms with their greater human and computer resources to do. The point is, these ideas are out there for someone able to bring them into play (plus something I mentioned on the podcast with Altucher is my own venture into sequencing and a form of minimization-on-small-data I have less objection to), so your putative smart cheater would have more to consider.