Module bandit 

Multi-armed bandits (Practical Statistics ch 3) — sequential experimentation that adapts: instead of a fixed-split A/B test, allocate more trials to the arms that look better, trading exploration against exploitation. Three classic policies: ε-greedy, UCB1, and Thompson sampling (Beta-Bernoulli). Kernel-class Divergent (the sampling/branch logic).

Structs

Bandit

A bandit over k arms.

Enums

Policy

Allocation policy.