{"data":{"kind":"file","path":"README.md","version_id":"hfe4vg96c2ex1en8xt9rzrlc","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":1595,"modified_at":"2026-08-20T13:56:37.076000","content_hash":"030ec349f6254056b5304ee72f67052e2c1019ae18a4dcbcc78cf8299ca232d2"},"entries":[],"content":"# Hanabi Cooperative Card Game Environment\n\nA verifiers environment that evaluates LLM reasoning through the lens of Hanabi — a cooperative card game where players give each other limited hints to play cards in the correct order.\n\n## Overview\n\nHanabi is unique among card games: you can see everyone's cards except your own. Players take turns giving hints (color or number) or playing cards. The goal is to play cards 1-5 in each of 5 colors. This environment tests:\n\n- **Deductive reasoning**: Inferring hidden information from hints\n- **Strategic planning**: Deciding when to hint, play, or discard\n- **Cooperative communication**: Maximizing information per hint\n- **Risk assessment**: Evaluating the safety of playing uncertain cards\n\n## Task Types\n\n1. **Hint Interpretation**: Given a game state and hint, determine which cards are indicated\n2. **Optimal Hint Selection**: Choose the best hint to give in a given situation\n3. **Play/Discard Decision**: Decide which card to play or discard based on available information\n4. **Game State Analysis**: Analyze a complete game and identify optimal strategies\n5. **Endgame Reasoning**: Make critical decisions when fuse tokens are low\n\n## Scoring\n\n- **Correctness**: Accuracy of hint interpretation and game decisions\n- **Reasoning Quality**: Depth and validity of strategic justification\n- **Risk Calibration**: Appropriate risk-taking given game state\n- **Cooperation Score**: How well the response maximizes team information\n\n## Usage\n\n```python\nimport verifiers as vf\n\nenv = vf.load_environment(\"tommii/hanabi-cooperative-card-game\")\n```\n","encoding":"utf-8","truncated":false,"total_bytes":1595},"status":null}