The gap between "decision computed from normalized facts" and "facts reflect the evidence"

[source-origin]

The trust model section is unusually honest, and it surfaces the seam I care about most.

You state the proofs cover deterministic policy evaluation over normalized facts and the stable outputs (decision, eligibility_class, reason_bits), while specifically not covering raw upstream KYC/AML/accreditation evidence truth — and that settlement is attestation-backed, with the API signing a contract-ready attestation bound to the decision.

So a relying party (or the settlement contract) can verify: given these normalized facts, the decision is correct, and the attestation is bound to that decision. My question is about the link upstream of that: raw evidence → normalized facts.

Is that normalization step covered by anything signed/auditable — i.e. is there a signed chain raw evidence → normalized facts → decision → on-chain action that a third party can walk end to end, or does the raw→normalized link rest on trusting the issuer/gateway? The distinction matters because a verifier can otherwise confirm that a wrong but faithfully-normalized fact set produced the correct decision, with the whole chain still verifying.

If the normalization is deliberately out of scope for now, what’s the intended anchor for it later (issuer-signed fact attestations, a separate proof program, …)?