Jackdaw
Bit-exact headless Balatro v1.0.1o simulator for reinforcement learning research.
Pages
- RL Project Guide — Observations, actions, masks, rewards, the Gymnasium wrapper (works with MaskablePPO) and the vector env
- Evaluating on Jackdaw Arena — The hosted benchmark: server-held seeds, a fixed budget per agent, verified and public results
- Python API — The Rust-backed
Game: actions, legal moves, views, previews, snapshots, GameBuilder, speed
- Lockstep against the live game —
jackdaw lockstep, divergence bundles, replay, the deviation catalog, and where the live game isn’t vanilla
- Validation & Scenarios — Validating the engine against live Balatro, writing your own test scenarios
- BalatroBot on the simulator —
jackdaw serve, SimBackend, and how the simulator maps onto BalatroBot’s API
- Snapshot schema —
jackdaw.snapshot/1, the save and comparison format
- Development — The toolchain, the everyday commands, game data and fixtures