MCTS 2048

Mesocosm typed environment: play 2048 on a 4×4 grid and merge tiles toward 512 within 35 moves.

• Env ID: mcts-2048
• Model (platform runs): openai/gpt-4o (multi-step; lower --episodes if rate-limited)
• Showcase: https://adityagupta7047.github.io/mcts-2048/ (after GitHub Pages deploy)

Local dev

python -m venv .venv
.venv\Scripts\activate   # Windows
pip install swecc-mesocosm numpy

python test_env.py       # determinism + policy checks
python adapter.py        # terminal 1
mesocosm run local       # terminal 2 (Ollama)

Submit

mesocosm auth login
mesocosm env submit --name "MCTS 2048" --github-url https://github.com/<user>/mcts-2048 --solo
mesocosm run create --domain <DOMAIN_ID> --vow-version 1.0.0 --model openai/gpt-4o --episodes 8 --visibility gallery_public --solo
mesocosm run export <RUN_ID> -o showcase/data/replay.json
python gen_sample_replay.py   # refreshes replay.js for file:// preview

Layout

benchanything.json   # manifest
env.py               # Game2048Env
adapter.py           # HTTP adapter
policies.py          # MCTS + heuristics for tests
test_env.py          # local validation
showcase/            # interactive replay UI