During each ranking pass, the ranker model generates per-item rationales explaining why each candidate ranks higher or lower than its neighbors, which are fed back to guide subsequent edit rounds and synthesize competitor narrative analyses.

flowchart TD
    A[Candidate Snippets] --> B[Ranker Model]
    B --> C[Median Rank Scoring]
    B --> D[Per-Item Rationales]
    D --> E[Feedback Loop]
    D --> F[Narrative Synthesis]
    E --> G[Ideator Prompting]
    G --> A
    F --> H[Competitor Analysis]