The Content Optimizer classifies snippet and page hypotheses into six structured mechanism classes—alignment, substance, architecture, style, framing, and proof—alongside a top-level escape class, assigning each claim a positional code (rfid) via an automated classifier. The acquisition engine uses these class labels to weight selection by exploration freshness before drawing Bayesian samples, while an automated custodian reviews unmapped claims in <class>.other buckets to discover and propose new domain-specific factors for human approval.