{"data":{"kind":"file","path":"README.md","version_id":"vgdlm1p5m21vxz57vi888zvq","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":1724,"modified_at":"2026-08-09T08:16:33.900000","content_hash":"84d46e7471ecb8843559cdeb3358da1ee5dbb675f01f192a62f42f7e9154850d"},"entries":[],"content":"# Wordle Hangman Advanced\n\nA multi-turn game environment that tests deduction, strategy, and information-theoretic reasoning through two classic word-guessing games.\n\n## Games\n\n### Wordle\nGuess a 5-letter word in 6 attempts. After each guess, receive color-coded feedback:\n- **GREEN**: Correct letter in correct position\n- **YELLOW**: Correct letter in wrong position  \n- **GRAY**: Letter not in the word\n\n### Hangman\nGuess a hidden word one letter at a time with limited wrong guesses. The model must deduce the word from partial reveals and wrong-guess feedback.\n\n## Difficulty Levels\n\n| Level | Game | Word Length | Max Guesses | Description |\n|-------|------|-------------|-------------|-------------|\n| 0 | Wordle | 5 | 6 | Standard Wordle |\n| 1 | Wordle | 5 | 5 | Hard mode Wordle |\n| 2 | Hangman | 4-6 | 6 | Classic Hangman |\n| 3 | Hangman | 7-9 | 5 | Long words, fewer guesses |\n| 4 | Mixed | 5-8 | 4-6 | Random game selection, tight limits |\n\n## Reward Functions\n\n- **wordle_win_rate**: Fraction of Wordle games solved within guess limit\n- **hangman_win_rate**: Fraction of Hangman games solved within guess limit  \n- **guess_efficiency**: Average score based on fewer guesses used (normalized 0-1)\n- **information_gain**: Measures quality of letter selection (common letters first)\n- **invalid_guess_penalty**: Penalizes malformed or repeated guesses\n\n## Training\n\nThe environment trains models to:\n- Apply optimal opening strategies (high-information first guesses)\n- Use feedback to constrain the solution space\n- Balance exploration vs exploitation in letter selection\n- Handle uncertainty and probabilistic reasoning\n\n## Evaluation\n\n```bash\nprime eval run baja/wordle-hangman-advanced -m Qwen/Qwen3.5-0.8B\n```\n","encoding":"utf-8","truncated":false,"total_bytes":1724},"status":null}