I looked at the paper.. They used only positions analyzable by 6 piece tablebases, which was a surprise, i.e. only positions with a maximum of 6 pieces, e.g. ea

I looked at the paper.. They used only positions analyzable by 6 piece tablebases, which was a surprise, i.e. only positions with a maximum of 6 pieces, e.g. each side has king and 2 pawns. The game starts with 32 pieces, so that means eliminating all the complex board positions I imagined a study of chess complexity would involve! So they're not so much studying chess proper, as endgames, and very minimal ones at that. They have an new definition of blunder - "a player has committed an error if their move worsens the mini-max value from their perspective. That is, the player had a forced win before making their move but now they don’t; or the player had a forced draw before making their move but now they don’t. ...we will refer to such an instance as a blunder." .. "Since we are interested in studying errors, we exclude all instances in which the player to move is in a theoretically losing position — where the opponent has a direct path to checkmate — because there are no blunders in losing positions (the minimax value of the position is already as bad as possible for the player to move)." "Our data comes from two large databases of recorded chess games. The first is a corpus of approximately 200 million games from the Free Internet Chess Server (FICS), where amateurs play each other on-line. The second is a corpus of approximately 1 million games played in international tournaments by the strongest players in the world. ... we focus on this large subset of the FICS data consisting exclusively of games with 3 minutes allocated to each side." I've played 10,000+ games on FICS, so I have some idea how it works. Players have a standard, blitz and lightning rating, all separate. 15 0 and slower counts as 'standard', anything faster down to 3 0 is blitz, anything faster than 3 0 is lightning. [0] 3 0 is a different world to 14 0, so treating those ratings as the same is already misleading - I often played 5 14 (i.e. 5 minutes on your clock to start, plus 14 seconds added on after each move), a bit faster than 5 15, which is counted for rating purposes as the same as 15 0. (It assumes a game goes for 40 moves, so 5 minutes + 15 seconds x 40 moves = 15 minutes.) So 5 14 is the slowest blitz speed. If I played someone rated the same as me whose rating mainly came from 3 0 games, they would be much stronger than me at 5 14. (Usually, on FICS, the faster the time control, the higher the rating. And obviously the more time you have, the better moves you make.) The point being - someone's blitz rating didnt mean they got it playing at 3 0. It's not their 3 0 rating. Also, FICS doesn't use Elo. (It uses the Glicko rating system[1]) The paper's authors seem unaware of that, or maybe I missed the part where they explained how they converted between FICS and Elo. As far as I know, it's not possible. There's not a linear relationship, or any simple conversion formula. From memory, lower rated FICS players have higher Elo, higher rated FICS players (over 2200) have lower Elo etc. But chess websites have a variety of rating systems, none of them compatible, commensurable or equivalent. Sadly, I kind of lost interest there. I meant to study the paper more thoroughly. But already that rating confusion seems a huge 'blunder'. [0] https://www.freechess.org/Help/HelpFiles/blitz.html https://www.freechess.org/Help/HelpFiles/blitz.html https://www.freechess.org/Help/HelpFiles/standard.html https://www.freechess.org/Help/HelpFiles/standard.html [1]https://en.wikipedia.org/wiki/Glicko_rating_system https://en.wikipedia.org/wiki/Glicko_rating_system