Claude vs DeepSeek chess, refereed by a script that drives terminal panes with Rex send + capture - ndedic/rex-chess-match ...

Max effort really is a different gear. On the same 30 chess puzzles Opus 5.5 went from 53% solved at medium to 88% at max, but thinking grew from about 800 to 9,500 tokens per answer. Would love to ...

Lichess turns a Stockfish eval into winning chances with 2/ (1+exp (−k·cp))−1 (k = 0.00368) and uses the drop in winning chances to mark inaccuracies, mistakes and blunders. This code fits k to real ...