The ranker compares candidate edits against competitor items across multiple judge samples to determine the median observed position, awarding a binary reward of one when rank improves and updating the underlying Beta distribution to guide future hypothesis selection.

flowchart LR
    A[Candidate Edit Generated] --> B[Ranker Multi-Sample Scoring]
    B --> C[Compute Median Rank]
    C --> D{Better than Best Rank?}
    D -- Yes --> E[Reward = 1, Update Best Rank]
    D -- No --> F[Reward = 0]
    E --> G[Update Beta: Alpha += 1]
    F --> H[Update Beta: Beta += 1]
    G --> I[Derive Claim Status]
    H --> I