The optimization loop seeds a Beta prior on each claim, uses Thompson Sampling to select the next hypothesis, lets the Ideator craft a revised snippet, and updates the belief distribution after the Ranker evaluates the competitive set.