{"data":{"kind":"file","path":"README.md","version_id":"kjlxue9jzan97w1j7vu9hiee","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":1361,"modified_at":"2026-03-30T10:19:17.768000","content_hash":"fe89bce4cbb00bde04ec9a8aaff135baefc61602a9a68718cfc862d545369160"},"entries":[],"content":"# Tower Defence\n\nPrime environment package for tower-defense training and evaluation.\n\n## Overview\n- Environment ID: `kbediako/tower_defence`\n- Python package: `tower_defence`\n- Primary wrapper: `actual_game_command`\n- Legacy wrapper: `macro_round`\n\n## Actual-Game Command Surface\nThe current primary route is offline single-decision replay on the multiplayer bridge command surface.\n\n- Decision kinds: `command` and `noop`\n- Command types: `place_tower`, `sell_tower`, `upgrade_tower`, `trigger_next_round`\n- Packaged replay corpora live under `src/prime_td_env/data/actual_game_command/`\n\n## Quickstart\nRun hosted training on the canonical config:\n\n```bash\nprime rl run configs/lab/prime-td-actual-game-command-hosted.toml\n```\n\nRun packaged replay evaluation locally:\n\n```bash\nPYTHONPATH=src python3 scripts/run_actual_game_command_eval.py --mode gold\n```\n\n## Environment Arguments\n- `wrapper`: `macro_round`, `actual_game_command`, or `benchmark_commands`\n- `dataset.examples_path` / `dataset.examples_paths`: select packaged replay data\n- `num_examples`, `seed_start`: control sampling\n- `reward_weights.*`: blend format, family, payload, intermission-progress, legality, and exact env reward\n\n## Notes\n- `macro_round` remains available for legacy historical runs.\n- `actual_game_command` is the maintained route for current hosted training and evaluation.\n","encoding":"utf-8","truncated":false,"total_bytes":1361},"status":null}