The system categorizes snippet and page optimization hypotheses into seven locked classes, assigning each factor a unique positional identifier (rfid) and allowing the Custodian agent to propose new property-scoped sub-factors from empirical clusters.

flowchart TD
    A[Taxonomy Architecture] --> B[1. Alignment]
    A --> C[2. Substance]
    A --> D[3. Architecture]
    A --> E[4. Style]
    A --> F[5. Framing]
    A --> G[6. Proof]
    A --> H[7. Other]
    C --> C1[2.1 Named Entities]
    C --> C2[2.4 Comparative Proposition]
    C --> C3["2.7@90 Custom Minted Factor"]