Impact summary
The shared evidence verifier now returns independently verified provider,
retirement, and assignment receipt sets. Assignment receipts replay against their
registered public review-readiness packets. A missing half receives no assignment
credit; a malformed pair is withheld with only the bounded surface/code diagnostic
`ASSIGNMENT_RECEIPT_INVALID`. Provider evidence remains available, preventing a
damaged reviewer handoff from hiding a valid trial. The readiness CLI's standalone
assignment loop and obsolete binding export are removed. Canonical readiness stays
at four provider surfaces, two assignments, two retirements, and zero advantages.
Ranked coupling and failure findings
- High — malformed diagnostics are console-only: canonical artifacts remain
privacy-minimal, but an operator reading only JSON cannot distinguish missing
assignment files from invalid ones.
- High — human returns remain absent: verified forms and isolated diagnostics
improve handoff reliability but still do not establish model value.
- Medium — assignment verification is generic rather than kind-dispatched:
this is correct for the shared v1 format, but a future assignment schema needs
an explicit version migration rather than permissive branching.
Actions
| Owner | Action and acceptance criteria | Validation |
|---|---|---|
| AI Reliability | Preserve privacy-safe invalid-assignment blocker codes in readiness JSON without paths or parser messages | Mutation test shows provider credit true, assignment false, bounded blocker present |
| AI Evaluation | Distribute both prepared offline forms to genuine independent reviewers and retain exact timed response exports | Two distinct pseudonymous complete responses per packet |
| Trust | Preserve packet hash, two distinct form digests, false authority, and no identity/candidate fields for every assignment schema version | Blind-review and evidence-verifier mutation suites |
| Maintainers | Remove any future duplicate assignment loop; all readiness assignment credit must originate in the shared verified set | Consumer search plus readiness CLI test |
Next-cycle hypothesis
Two completed response receipts can be verified and routed with the same isolation
property: a malformed review receipt must not erase provider or assignment credit,
must not expose reviewer data, and must never count as completed review. Falsify
with packet, score, reviewer-code, and authority mutations.