A chess engine
that learns.
Take a seat. The house engine opens from a book of master games, plays the middlegame with a network trained on real chess, and converts endgames with exact solvers. Every night it trains; Stockfish grades it, so the rating on the plate is honest.
From the book
The first moves come from an opening book learned from thousands of 2000+ rated games. Sound, varied, principled.
From the network
A residual net cloned from real games, then refined by nightly self-play. A new generation is kept only if it beats the champion.
From the tables
Solved endgames are played perfectly with exact dynamic programming. Won positions get converted, not fumbled.
Stockfish grades it.
It never plays for it.
Every move belongs to the trained agents. Stockfish sits outside the game as the arbiter: throttled to known ratings, it plays grading matches and signs off the calibrated Elo you see. No self-reported numbers.
Every training generation is an entry: its rating, and whether it earned the table. A generation is promoted only if it beats the sitting champion.
The scorebook opens when the backend wakes.