Research notes on chess analytics, rating mappings, benchmark construction, and data-driven improvement methods used by Elo+Chess.
A methodology note explaining how Elo+Chess builds stratified Lichess benchmark curves, player-history percentile targets, and production report graphics from large move-state and player-game samples.
Read paperA methods note describing the Elo+Chess same-username rating-mapping study, the modal fit motivation, filtering criteria, and current Lichess-to-Chess.com conversion estimates by game type.
Read paperA public Medium article sharing Elo+Chess research observations and charts for a wider chess audience.
Read on MediumAn observational Lichess.org study of more than one billion public games testing whether 1min bullet and 3min blitz volume is associated with slower 10min rapid improvement for beginner and club-level players.
Read articleA low-key research note on Lichess 10+0 rapid resignation-like endings. In this dataset, well-sampled rating bands from under 800 to nearly 1800 typically resign around a 7-pawn Stockfish disadvantage.
Read articleA Lichess study comparing how many pieces remain on the board at valid checkmate and engine-confirmed resignation-like endings across Bullet, Blitz, 10min Rapid, and >10min Rapid.
Read studyStatic research gallery showing the average move number before each original pawn is first touched, split by game type, color, pawn file, and Elo bucket.
Explore galleryStatic research gallery for White f3 and Black f6 rates, including before castling, before development completion, and associated win-rate views by Elo bucket.
Explore galleryStatic research gallery showing the average row where pawns are captured, how often promoted pieces are captured immediately, and how often captured promotions produce no net material gain.
Explore gallery