Concepts

Concept pages explain why reinfors divides responsibilities as it does. They are optional context for the first example and useful when choosing an experiment architecture.

• Architecture maps games, encoders, policies, learners, solvers, Engine, and Env onto their Rust and Python responsibilities.
• Sampling and training follows one collection round, explains pooled inference, and connects n_games and search budgets to hardware utilization.
• Engine collection internals shows how one collection window executes across threads: the zero-copy observation path, the inference cache, and liveness.

When the model is clear, put it into practice with the training guide.