Govern

Sign-off is the moment a report leaves the working state and becomes a defensible artefact. The queue routes each report to the right reviewer, surfaces the evidence bundle and decision history, and locks the report's content into the audit trail at the moment of approval.

Sign-off is not a rubber stamp. The reviewer can request changes, return the report to the assessor, or escalate to an integration session if the dissent is material. The platform tracks the SLA; signed-off reports carry the reviewer's identity, the timestamp, and the hash.

Corrections are different from contests. A correction is a factual fix — name spelled wrong, transcript misattributed, the wrong example linked to a rating. The pathway is lighter than a contest because the question is not whether the rating is right; it is whether the record is accurate.

Every correction is logged. The original is preserved alongside the corrected version; the audit trail makes clear what changed and why.

The fairness surface watches the platform's behaviour across demographic attributes — disparate impact at the four-fifths rule, equalised odds across true-positive and false-positive rates, calibration against the underlying distribution, and counterfactual fairness through attribute-swap audits.

Each metric ships with an anchored plain-language band — typical, worth a look, investigate — and a one-paragraph explanation of what causes drift. The platform's posture is fairness as a posture pointed at itself, not at the practitioner. Alerts route into an open-investigation surface with a recommended response.

Seven sector packs flex tone, palette, distress thresholds, consent shape, audit posture, and example phrasings — government (Australian and international), corporate, community, healthcare, education, finance. The Indigenous-led overlay adds Acknowledgement of Country, relational pacing, and additional cultural primitives.

Sector posture is one choice per deployment. The setting flows through every other surface — the welcome the participant sees, the consent layers, the safety thresholds, the audit-ready posture of defensibility bundles. Practitioners do not author sector copy ad hoc; the pack is platform-managed.

Every consequential action — consent change, role change, programme state transition, sign-off, key admin move — lands in the audit log. Entries are hash-chained so any post-hoc edit is detectable; the log is filterable, exportable, and retained for the period the organisation's policy specifies.

The audit log is not marketing language. It is the artefact an internal auditor or an external regulator reads when they want to know what actually happened, in what order, by whom.

The counterfactual audit takes an evidence bundle, swaps a demographic attribute, and re-runs the rating. The surface reports the cases where the rating moves. A rating that moves under a counterfactual swap is the rating to look at — not the rating to defend.

Counterfactual audits ship on a routine cadence. The surface is built for cohort review, not for individual remediation. The platform's intent is to catch attribute-driven drift before it lands as a contest.