First game: Mortal Kombat II (Genesis), Liu Kang vs the CPU. Models read game memory, not pixels.

Edition mk2-liukang-v1. Prompt hash (SHA-256 of the edition file): loading

Policy Table

The model answers 8 situation questions once per fight, before play. Rank: mean damage dealt on the 15-fight VeryHard ladder, 10 repeats. Latency: n/a.

Live

The model answers the current situation every 0.12 s, at most 3 requests in flight. Rank: replayed mean damage dealt. Latency is self-reported and is not a rank key.

Method

• Same save states, same 17 moves, same prompt, same seeds for every row.
• Rank is mean damage dealt, high to low. Wins are reported. Wins are not the sort key.
• script is the rules-bot floor row. random picks uniformly from the 17 moves.
• A different model version or track is a different row. A different prompt or temperature is a new edition.
• Replay checks the game outcome. It does not re-call the model. The model name is attested by the submitter.
• Smoke rows (one fight, one round) are shown in grey, below the official rows. One fight is not the ladder.

Submit

1. Install your own ROM: uv run mk2_clef.py --rom <your file>. Nobody ships a ROM.
2. Write a Policy class (see fightbench.py).
3. Policy Table: uv run fightbench.py table --policy my.py:MyPolicy. Live: uv run fightbench.py live --policy my.py:MyPolicy.
4. Commit only the file in submissions/ on a branch and open a pull request.
5. The maintainer runs uv run fightbench.py replay submissions/<file>.json. Only replayed results reach this board.

Code: github.com/aipdv/fightbench