The Bayesian optimization loop operates by seeding distinct claims against competitor content, selecting top candidates via a Bayesian acquisition strategy, generating line-level edits, and updating parameter posteriors through multi-sample re-ranking across rounds.

sequenceDiagram
    autonumber
    participant S as Seeder
    participant B as Bayesian State
    participant A as Acquisition Loop
    participant I as Ideator
    participant R as Ranker / Judge

    S->>B: Seed initial claims & Beta priors
    loop Each Round
        B->>A: Sample candidate claims
        A->>I: Pick winning claim
        I->>I: Generate patch edit on page
        I->>R: Submit modified page & competitor set
        R->>R: N-sample median re-ranking
        R->>B: Reward: 1 if improved, 0 otherwise
        B->>B: Update Beta(alpha, beta) posterior
    end