Skip to content

Latest commit

 

History

History
31 lines (24 loc) · 1.5 KB

File metadata and controls

31 lines (24 loc) · 1.5 KB

Colver Documentation

Belote Contrée engine + RL training stack.

Top-level

  • ARCHITECTURE.md — workspace layout, key subsystems, observation formats
  • RULES.md — game rules summary
  • BENCH.md — performance benchmarks
  • arena_results.mdglobal arena leaderboard (the main eval metric)
  • règles officielles belote contrée.pdf — official FFB rules

Subsystems

  • bid/ — bidding strategies, bid NN training, reward studies, interpretability
  • play/ — play methods (DD solver, IS-DD, DMC, IS-MCTS), play NN training
  • belief/ — belief models for hidden card inference
  • data_gen/ — pool generation, enrichment methods, replay formats
  • training/ — training pipelines and how to invoke them
  • superpowers/ — agent-driven plans and design specs

Quick links by topic

Topic Main doc
Latest bid champion bid/reward_studies/v3_max_signal_results.md
Reward signal study bid/reward_studies/v3_reward_study.md
Distilled bid rules (XGBoost) bid/interpretability/bid_rules_xgb.md
Triforge joint training play/experiments/triforge.md
Belief net + IS-DD belief/bis_dd.md
Training commands training/overview.md