Chess Logic

reasoning by relaxation, not search

No tree, no minimax. Every legal move gets a belief from transparent chess logic, then the beliefs relax against each other — a move gains support when other good moves share its plan. Watch the beliefs argue until one wins.

A generative interpretability object: the same relaxation-labeling engine as the sims, applied to move selection. The priors are biased by a chess_mind style fingerprint, so it plays like the seed player — it hunts kings and dabs the queen. It won't outcalculate Stockfish; it will tell you why it moved.

← Cognitive Fingerprints · the grown Brain →

A chess engine that reasons by relaxation, not search

challenge accepted
No search tree. Just legal moves, logical priors, and a rule that lets beliefs argue.
mind = blown
Rosenfeld–Hummel–Zucker relaxation: a move gains belief when other good moves share its plan.
mind = blown
It converges on Qf6+ — and can name the plan and the moves that reinforced it.