The Bayesian optimization loop operates by generating candidate hypotheses, selecting one via two-stage acquisition selection, applying targeted patch edits to the content, and submitting the resulting variant to a re-ranking engine to measure rank movement and compute posterior updates.
flowchart LR
A[Seeded Hypotheses] --> B[Two-Stage Acquisition: Sample Claim]
B --> C[Ideator: Apply Patch Edits]
C --> D[Ranker: N-Sample Evaluation]
D --> E{Rank Improved?}
E -- Yes --> F[Reward=1: Update Posterior & Set New Best]
E -- No --> G[Reward=0: Update Posterior]
F --> H{Objective Met or Budget End?}
G --> H
H -- No --> B
H -- Yes --> I[Winning Content & Content Brief]