Belot card game engine written in C#
Belot (Bridge-Belot or Belote) is a 32-card, trick-taking game popular in Bulgaria, France, Armenia, Croatia, Cyprus, Greece, Moldova, North Macedonia and also in Saudi Arabia.
dotnet add package BelotGameEngine --version 2.0.0The NuGet package contains only the managed Belot.Engine library for .NET Standard 2.0,
with no bots, neural weights, UI or third-party runtime dependencies. See the
engine README for the BelotMatch API and migration from 1.1.
Publishing instructions describe the trusted GitHub workflow.
src/UI/Belot.UI is a .NET MAUI app for Android and Windows, in English and Bulgarian: you play
South, and pick the level of your partner and of each rival, from a random player to the neural
player ClaudePlayerNeural. It keeps your rating, your game history and your record against each
level on the device.
SmartPlayer: hand-written rules, improved change by change and measured in ELO.ClaudePlayerIsmcts: Information Set Monte Carlo Tree Search, +375 ELO over SmartPlayer.ClaudePlayerNeural: neural networks trained by reinforcement learning in self-play, which value every bid and every card. Alone they tie ClaudePlayerIsmcts at about 10 µs a decision; with a small search on top (SearchDeals) they beat it. See NEURAL_NETWORK.md.- The app's Master (
ClaudePlayerProfiles.CreateMaster): the networks with search in every trick and a strong human player's technique and conventions (it bids naturally, keeps its high cards, signals to its partner and reads its partner's signals); it beats every other bot here. See HUMAN_PLAY.md. ClaudePlayerHeuristic: the written advice of the belot.bg academy and other Bulgarian sources as rules, no networks, with the last five tricks played out exactly over the deals the play allows (under 1 ms a card). It beats SmartPlayer 87.7% and the 2001 Belot 2.06 72.3%, and comes within 33 ELO of ClaudePlayerIsmcts. See HEURISTIC_PLAYER.md.
