Verification Playbook: Make Trust Model Inspectable #12
Description
Priority: Critical
Description
The documentation asserts the trust model but doesn't demonstrate it. Claims include:
- "Determinism Guarantees" (precision rounding, stateless, pinned PostGIS)
- "TEE-held key" - only code running inside the TEE can produce valid signatures
- "EigenCompute provides hardware attestation of execution"
However, the documentation doesn't explain what a verifier (developer, user, auditor) can actually check:
- What evidence do they receive about the TEE measurement/attestation?
- What is the verification procedure (inputs, outputs, signatures, attestation report)?
- What parts are roadmap vs shipping?
Without this, you're asking people to trust marketing statements. The story is plausible but not inspectable.
Fix Direction
Add a concrete "Verification Playbook" section with:
- What is signed (exact bytes / struct encoding)
- How to verify
attestation.attesterand the delegated signature path - What a TEE attestation artifact looks like in EigenCompute
- Where to fetch the attestation artifact
- How it binds to the signing key
- Clear "MVP limitations" (e.g., if TEE attestation is not exposed yet, say so bluntly)