For each claim, Thompson sampling draws a random sample \(\theta_i\) from a Beta distribution parameterized by the claim's current \(\alpha\) and \(\beta\) values, selecting the claim that produces the highest sampled value to test in the next optimization round.

flowchart TD
    A[Start Selection] --> B[Retrieve Hypotheses in Class]
    B --> C["Sample theta_i ~ Beta(alpha_i, beta_i)"]
    C --> D[Find argmax theta_i]
    D --> E[Select Winning Claim for Test Round]