The Bayesian Content Optimizer seeds ranking hypotheses, selects them via a Bayesian acquisition method, generates rewritten snippet variants with an ideator, and iteratively scores them with a multi-sample LLM ranker to update posterior beliefs until the snippet converges to rank one.
flowchart TD
A[Seed Initial Claims] --> B[Two-Stage Acquisition]
B --> C[Ideator Crafts Variant]
C --> D[Multi-Sample Ranker]
D --> E{Improved Rank?}
E -- Yes --> F[Reward = 1, Update Beta Prior, Set New Best]
E -- No --> G[Reward = 0, Update Beta Prior]
F --> H{Rank 1 or Budget Met?}
G --> H
H -- No --> B
H -- Yes --> I[Output Winning Snippet and Narrative Brief]Referenced by