The part I'd poke at is the claim that the verifier retains no personal fields locally, because the proof itself can still become a stable handle if the portal logs timestamps, IPs, or repeated eligibility responses. In a grant flow, that kind of linkage is often enough to re-identify the student even when the DB row is gone. ML-DSA-65 for attestation is a sensible post-quantum choice, but it also bumps signature size and verification cost compared with a plain Ed25519-style setup, so the edge box still needs to be sized for bursty intake rather than just the happy path.