{"data":{"kind":"file","path":"README.md","version_id":"dfe2zlx6jlqjw2nqt8252xxi","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":2005,"modified_at":"2026-08-09T05:37:29.703000","content_hash":"4e710e180df7a72e9d2794dab5ccbc3685bc72b0aafe84b230b2e5c038c51ede"},"entries":[],"content":"# Sliding Puzzle Environment\n\nA verifiers environment that tests LLM spatial reasoning and planning through sliding puzzles (15-puzzle and Klotski variants).\n\n## Overview\n\nThe model receives a scrambled sliding puzzle grid and must output the sequence of moves (U/D/L/R) to reach the goal state. Puzzles are generated by shuffling a solved grid, with A* search providing optimal solutions as ground truth.\n\n## Puzzle Types\n\n| Level | Grid  | Shuffles | Name        |\n|-------|-------|----------|-------------|\n| 0     | 3x3   | 5        | 3x3 Easy    |\n| 1     | 3x3   | 15       | 3x3 Medium  |\n| 2     | 3x3   | 30       | 3x3 Hard    |\n| 3     | 4x4   | 10       | 4x4 Easy    |\n| 4     | 4x4   | 25       | 4x4 Medium  |\n| 5     | 4x4   | 50       | 4x4 Hard    |\n\n## Reward Functions\n\n| Function                | Weight | Description                                           |\n|-------------------------|--------|-------------------------------------------------------|\n| Solution Correctness    | 3.0    | Does the move sequence actually solve the puzzle?     |\n| Move Efficiency         | 2.0    | How close to optimal move count (sigmoid-smoothed)    |\n| Move Validity           | 1.5    | Fraction of valid moves before first invalid move     |\n| Format Compliance       | 0.5    | Output uses expected format (MOVES: marker, etc.)     |\n| Reasoning Quality       | 0.5    | Step-by-step reasoning about tile movements           |\n\n## Configuration\n\n```python\nenv = load_environment(\n    num_samples=30,   # Number of puzzles to generate\n    max_level=4,      # Maximum difficulty level (0-5)\n    seed=42,          # Random seed\n)\n```\n\n## Example Prompt\n\n```\nSolve this 3x3 sliding puzzle.\n\nCurrent state:\n  1   2   3\n  4   _   5\n  7   8   6\n\nGoal state:\n  1   2   3\n  4   5   6\n  7   8   _\n```\n\nExpected output:\n```\nI need to move tiles into position step by step.\n\n1. Move R (blank goes right, tile 5 slides left)\n2. Move D (blank goes down, tile 8 slides up)\n\nMOVES: R D\n```\n\n## License\n\nMIT\n","encoding":"utf-8","truncated":false,"total_bytes":2005},"status":null}