{"data":{"kind":"file","path":"README.md","version_id":"qluin5h4b1f7cxwmsa2dk6cf","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":929,"modified_at":"2026-08-04T07:20:56.229000","content_hash":"20fe0e67fcd5c138551c64a55a2dd880a50099b66ed6ba81f6f556e90529ecdb"},"entries":[],"content":"# Tetris RL Eval\n\nEvaluate LLM agents on Tetris reinforcement learning tasks. The agent must generate working Python strategies that play Tetris, then the environment simulates the game and scores performance.\n\n## Tasks\n\n- **Strategy Design**: Agent writes a Tetris strategy function that selects the best piece placement given the current board state and piece\n- **RL Implementation**: Agent implements value functions, Q-learning policies, or heuristic evaluators for Tetris\n- **Code Quality**: Evaluated on correctness, efficiency, and robustness of generated code\n\n## Scoring\n\n- **Correctness** (0-1): Code executes without errors and produces valid moves\n- **Performance** (0-1): Lines cleared / target lines across multiple game seeds\n- **Code Quality** (0-1): Proper structure, edge case handling, algorithm sophistication\n\n## Usage\n\n```python\nimport verifiers as vf\nenv = vf.load_environment(\"thanos/tetris-rl-eval\")\n```\n","encoding":"utf-8","truncated":false,"total_bytes":929},"status":null}