{"data":{"kind":"file","path":"README.md","version_id":"lcpze4xs6j7gcdgavxv13oys","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":2132,"modified_at":"2026-08-09T07:47:33.589000","content_hash":"cc656202225c6aa69a958c016d0f3bcfbc0d401a0b47a0d68e7eb4603b4c0880"},"entries":[],"content":"# Sea Battle Agent\n\nA multi-turn Battleship (Sea Battle) environment for training and evaluating strategic game-playing agents.\n\n## Overview\n\nThe agent plays Battleship on grids ranging from 6x6 to 10x10, with ship fleets of varying complexity (1-5 ships). The environment tests:\n\n- **Probabilistic reasoning**: Using hit/miss patterns to compute optimal target probabilities\n- **Hunt/Target strategy**: Systematic search (hunt mode) vs focused attack on hit clusters (target mode)\n- **Ship placement logic**: Understanding how remaining ships can fit on the board\n- **Adaptive gameplay**: Adjusting strategy based on partial information\n\n## Difficulty Levels\n\n| Level | Board | Ships | Description |\n|-------|-------|-------|-------------|\n| 0 | 6x6 | 1 | Beginner — single destroyer |\n| 1 | 8x8 | 2 | Easy — destroyer + submarine |\n| 2 | 10x10 | 3 | Medium — add cruiser |\n| 3 | 10x10 | 4 | Hard — add battleship |\n| 4 | 10x10 | 5 | Expert — full fleet including carrier |\n\n## Reward Functions\n\n| Reward | Weight | Type | Description |\n|--------|--------|------|-------------|\n| `move_accuracy_reward` | 0.35 | Proxy | Whether the chosen coordinate matches optimal probability-based move |\n| `strategic_reasoning_reward` | 0.30 | True | Quality of explanation: probability concepts, coordinate analysis, ship logic |\n| `hunt_target_strategy_reward` | 0.20 | True | Demonstrates hunt/target strategy awareness |\n| `move_format_reward` | 0.15 | Proxy | Proper `<move>row,col</move>` format |\n\n## Usage\n\n```python\nimport verifiers as vf\n\nenv = vf.load_environment(\"sea-battle-agent\", level=2, num_examples=20)\n```\n\n## Configuration\n\n- `level` (int, 0-4): Difficulty level. Default: 1\n- `num_examples` (int): Number of scenarios in dataset. Default: 30\n- `max_turns` (int): Max turns per multi-turn game. Default: 8\n\n## Game Mechanics\n\n1. Agent receives a partially-played game state with targeting grid\n2. Agent analyzes patterns and explains reasoning\n3. Agent outputs move as `<move>row,col</move>`\n4. Environment returns hit/miss/sunk result with updated grid\n5. Game continues until all ships sunk or max turns reached\n","encoding":"utf-8","truncated":false,"total_bytes":2132},"status":null}