{"data":{"kind":"file","path":"README.md","version_id":"bioljv5wu06ygdhva5egj1d3","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":1464,"modified_at":"2026-08-03T13:46:58.193000","content_hash":"7d413229a68b69a8c1f9131355c7792efeb2b9492c74fa6278499c4099fdfbde"},"entries":[],"content":"# Puzzle Solver Environment\n\nA text-based puzzle-solving environment for evaluating LLM strategic planning and logical deduction capabilities.\n\n## Puzzles\n\n### 2048 (5 scenarios)\n- Pre-configured board states requiring strategic analysis\n- Evaluates: tile merging strategy, corner control, monotonic ordering\n- Open-ended: scored on reasoning quality and planning depth\n\n### Sudoku (4 puzzles, Easy to Expert)\n- Puzzles ranging from ~35 clues (easy) to ~22 clues (expert)\n- Evaluates: logical deduction, technique application, systematic solving\n- Scored on correctness of complete solution\n\n## Scoring\n\n| Metric | Weight | Description |\n|--------|--------|-------------|\n| Answer Format | 0.20 | Clear structure, step-by-step reasoning |\n| Logic Quality | 0.30 | Planning, alternatives, domain reasoning |\n| Completeness | 0.20 | Fully addresses the question |\n| Sudoku Accuracy | 0.30 | Correctness of solved grids (neutral for 2048) |\n\n## Usage\n\n```python\nimport verifiers as vf\n\nenv = vf.load_environment(\"basilaa/puzzle-solver\")\n# or\nfrom puzzle_solver import load_environment\nenv = load_environment()\n```\n\n## Configuration\n\n- `n`: Number of puzzles to include (0 = all 9 puzzles)\n\n## Design Notes\n\n- 2048 scenarios are open-ended strategy questions (no single correct move)\n- Sudoku puzzles have verified unique solutions\n- Sudoku accuracy uses full grid validation (rows, columns, 3x3 boxes)\n- Scoring rewards depth of reasoning over speed\n\n## License\n\nMIT\n","encoding":"utf-8","truncated":false,"total_bytes":1464},"status":null}