The Bayesian optimization loop elicits testable hypothesis claims, selects them using Thompson sampling and class-level freshness, crafts candidate snippet or page edits, and evaluates the resulting rank changes across multiple judge samples until the target reaches rank one or budget is exhausted.