lichess.org
Donate

Fog of war chess against an open-source engine

I wrote the engine below and run the site it plays on.

Fog of war has come up here more than once. In 2015, lila's answer (issue 441) named the hard parts: dark chess would be hard to keep free of cheating, has no spectacle, and "requires a complex referee-ing system." The referee is the part I built. On Mistboard the server keeps the real board and sends each player, engine and spectator only what that seat can see, and a finished game replays in full.

The engine, Misty, is GPL, Python and Rust. It follows Obscuro (Zhang and Sandholm), the first program to play the game at a superhuman level. It tracks the boards that fit what it has seen, up to a cap, and runs counterfactual regret minimization, the poker method, over a sample of them. Misty isn't at Obscuro's level, and games against people show where it falls short.

If you play fog on chess.com: what kind of position do you expect an engine to misjudge?

https://brianhliou.com/posts/misty-fog-of-war-engine/

I wrote the engine below and run the site it plays on. Fog of war has come up here more than once. In 2015, lila's answer (issue 441) named the hard parts: dark chess would be hard to keep free of cheating, has no spectacle, and "requires a complex referee-ing system." The referee is the part I built. On Mistboard the server keeps the real board and sends each player, engine and spectator only what that seat can see, and a finished game replays in full. The engine, Misty, is GPL, Python and Rust. It follows Obscuro (Zhang and Sandholm), the first program to play the game at a superhuman level. It tracks the boards that fit what it has seen, up to a cap, and runs counterfactual regret minimization, the poker method, over a sample of them. Misty isn't at Obscuro's level, and games against people show where it falls short. If you play fog on chess.com: what kind of position do you expect an engine to misjudge? https://brianhliou.com/posts/misty-fog-of-war-engine/