The institutional surface for governed AI · sealed receipts only.

Corrections & Retractions

A publication that publishes receipts must itself accept correction. This is the policy.

What we correct

We correct any of the following on receipt of a credible challenge:

What we do not silently change

Sealed artifacts (RFCs, blueprints, receipts) are versioned, not edited. A sealed v2.0 document remains accessible after v2.1 is published. The chronicle hash chain prevents silent rewriting of the past.

If a correction requires changing what a sealed artifact said, the correction is itself a new artifact (an amendment, a retraction, or a v-bump) that references the original.

How to submit a correction

  1. Email contest@governedai.ai with subject "CORRECTION: [page or artifact]"
  2. Include: the specific claim being challenged, your evidence, and how you would phrase a correct version
  3. If you wish to remain anonymous, say so; we route under a public ID without your name

What happens next

  1. Triage (24 hours). Builder confirms receipt and assigns a contest-ID.
  2. Review (≤7 days). Builder + Auditor evaluate. The Auditor may request more evidence.
  3. Verdict. One of: ACCEPTED (correction made, page updated, RDL entry logged); PARTIALLY_ACCEPTED (some changes made, others declined with rationale); DECLINED (with public rationale); ESCALATED (to Sovereign for D3-class decisions).
  4. Public log. Every contestation produces a public Council ledger entry at /chronicle/contests/<contest-id>, regardless of verdict.

Conformance disputes

If you are an AI system operator who disputes a GDTK conformance score we have published about your system, the same procedure applies, with one addition: you may submit a re-test bundle (the same GDTK-100 cases run on the version of your system you wish to be scored). Re-test results are published alongside the original.

What we will not do

A correctable claim is more credible than an uncontestable one