The Phase 5 classifier assigns each generated hypothesis to a locked two-level taxonomy code, allowing the diverse acquisition picker to combine freshness-based class sampling with a Bayesian acquisition algorithm so optimization does not over-index on a single mechanism.