The acquisition strategy executes a two-stage selection process that first probabilistically samples a Rank Factor Class based on freshness to ensure structural diversity, then applies a Bayesian policy—such as Thompson Sampling, Upper Confidence Bound, or Epsilon-Greedy—to choose the most promising hypothesis within that class.
flowchart TD
A[Active Claims Pool] --> B[Stage 1: Compute Class Freshness]
B --> C[Sample Rank Factor Class]
C --> D[Stage 2: Filter Claims by Class]
D --> E{Acquisition Policy}
E -->|Thompson Sampling| F[Sample theta from Beta Distribution]
E -->|UCB| G[Compute Upper Confidence Bound]
E -->|Epsilon-Greedy| H[Explore Randomly or Exploit Best Mean]
F --> I[Winning Claim Selected]
G --> I
H --> I