{"data":{"kind":"file","path":"README.md","version_id":"c615s06u3btd864c4c4xd9mn","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":2133,"modified_at":"2026-08-04T04:11:06.864000","content_hash":"f41b4b51b6a85252ea24f317e756b56b452f00c328f69a61b39137403bd38008"},"entries":[],"content":"# Chess & Go Variants (Mini-Board)\n\nEvaluate LLM reasoning on miniature versions of Chess and Go — classic strategy games reduced to smaller boards where every move matters more.\n\n## Overview\n\nThis environment tests spatial reasoning, tactical calculation, and strategic planning on mini-board variants:\n\n- **Mini Chess (5×5 / 6×6)** — Reduced piece sets, faster games, but full tactical depth\n- **Mini Go (5×5 / 7×7 / 9×9)** — Territory control on smaller grids with all standard rules\n\n## Task Types\n\n| Task | Description | Turn Type |\n|------|-------------|-----------|\n| **Best Move** | Given a board position, find the optimal move | Single |\n| **Tactical Puzzle** | Solve checkmate-in-N or capture sequences | Single |\n| **Board Evaluation** | Assess who is winning and by how much | Single |\n| **Game Play** | Play a complete mini-game against a rule-based opponent | Multi-Turn |\n\n## Board Representations\n\n### Mini Chess (5×5)\n```\nr n b q k\np p p p p\n. . . . .\nP P P P P\nR N B Q K\n```\nPieces: R/Rook, N/Knight, B/Bishop, Q/Queen, K/King, P/Pawn (uppercase=White)\n\n### Mini Go (5×5 / 7×7)\n```\n. . . . .\n. . . . .\n. . . . .\n. . . . .\n. . . . .\n```\nStones: B=Black, W=White, .=Empty. Standard Go rules (liberties, ko, suicide forbidden).\n\n## Reward Functions\n\n- **format_compliance** — Response follows required notation format\n- **move_legality** — Proposed move is legal on the given board\n- **tactical_accuracy** — Move matches the optimal solution (for puzzles)\n- **evaluation_accuracy** — Board evaluation matches ground truth (for eval tasks)\n\n## Scoring\n\nEach reward function returns 0.0-1.0. Weights:\n- format_compliance: 0.2\n- move_legality: 0.3\n- tactical_accuracy: 0.5 (for puzzle tasks)\n- evaluation_accuracy: 0.5 (for evaluation tasks)\n\n## Usage\n\n```bash\nprime eval run haha/chess-go-variants -m meta-llama/Llama-3.2-1B-Instruct\n```\n\n## Environment Args\n\n- `difficulty` (0-2): 0=easy, 1=medium, 2=hard (default: 1)\n- `game_type` (\"chess\", \"go\", \"mixed\"): which game to focus on (default: \"mixed\")\n- `board_size` (\"small\", \"medium\", \"all\"): board size filter (default: \"all\")\n\n## License\n\nMIT\n","encoding":"utf-8","truncated":false,"total_bytes":2133},"status":null}