Stage 2 evaluates candidate claims within the chosen class by sampling reward probabilities using Thompson sampling, upper confidence bound, or epsilon-greedy selection to pick the highest-potential hypothesis for testing.