跳转至

IOT-T051 Agent Advisory Review

Scope: M2 human review preparation. This document is an agent advisory review only. It is not a HUMAN_APPROVED review record, does not prove factual correctness, and must not be used to promote source_status or review_status.

Decision

No blocking findings for preparing a human review packet.

The current M2 state is correctly parked at PARKED_HUMAN_EVIDENCE: structural coverage and deployed-site acceptance are documented, while factual claim verification and independent human approval remain missing.

Evidence Checked

  • README.md and ROADMAP.md state that AI assistance cannot grant VERIFIED or HUMAN_APPROVED.
  • docs/progress.md states that M2 remains blocked on CONTENT_AUTHOR, FACT_AUDITOR, locked critical_claim_ids, CLAIM_VERIFICATION, and independent human approver evidence.
  • data/content-inventory.json reports source_audited_files=0, PARTIAL=0, and VERIFIED=0.
  • data/deploy-acceptance.yml records Pages acceptance for the earlier deployed SHA, but explicitly keeps trust counts at zero.
  • data/trust-authorities.yml contains structural-auditor authority only; it does not authorize a content author, fact auditor, or human approver.

Boundary Findings

  1. Agent review cannot substitute for human review.

  2. Required next evidence: a real human reviewer identity in the authority registry and an independent review record that binds the current body hash.

  3. Structural audits cannot substitute for claim verification.

  4. Required next evidence: at least one CLAIM_VERIFICATION source audit covering locked critical claims.

  5. IN_REVIEW legacy state cannot substitute for approval.

  6. Required next evidence: an evidence-bound review record; current review_records=0.

A human reviewer should only approve after all of the following are true:

  • A target content item is selected and its current body_sha256 is frozen.
  • CONTENT_AUTHOR authority exists for the content item.
  • critical_claim_ids are locked before fact review starts.
  • A FACT_AUDITOR independently verifies the selected claims through CLAIM_VERIFICATION.
  • A separate HUMAN approver reviews the verified evidence and writes a review record that references the active audit IDs.

Until those conditions exist, M2 should remain PARKED_HUMAN_EVIDENCE.