Model Review Log
This page is the record book of the scheduled reviews of the AI and machine learning components running on the platform. Each completed review lands here as a dated entry.
The obligation to conduct these reviews is undertaken in the Responsible AI Policy, and how they are conducted is defined in the Model Review Policy. This page is the evidence that the obligation is met.
No scheduled review has been completed yet. The cadence begins on the policy's effective date, 20 September 2026; the first review falls on 20 December 2026. Model and measurement audits were performed before that date, but as the need arose rather than on a schedule. We do not present that history as scheduled.
Schedule
| Type | Frequency | Scope |
|---|---|---|
| Model review | Quarterly | All models running in production and their decision authority |
| Policy review | Annually | The Responsible AI Policy itself |
| Event-driven review | On trigger | When a model's decision authority is widened or a new AI surface is added |
What each review covers
| Topic | What is examined |
|---|---|
| Inventory | Models running in production, their versions and their decision authority |
| Measurement | Shadow and gradual-rollout reports, sample sufficiency, confidence intervals |
| Explainability | A sample of decision rationales and their consistency with the decision |
| Incidents | Threshold alerts, rollbacks and the responses to them |
| Open items | The status of gaps carried over from the previous review |
Records
| Date | Type | Status | Outcome |
|---|---|---|---|
| 20 December 2026 | Model review | Scheduled | — |
| 20 March 2027 | Model review | Scheduled | — |
| 20 June 2027 | Model review | Scheduled | — |
| 20 September 2027 | Model review · Policy review | Scheduled | — |
When a review is completed, its row is updated to Completed and a summary is added beneath it: the model versions examined, the open items identified and those closed.
Requesting the detailed record
The summary in the table is public. The detailed record of a review — the measurement reports examined, the sampling results and the open-item list — is shared under NDA as part of a vendor assessment.
To request it: [email protected]
What this does not replace
Review does not replace monitoring. Model monitoring is continuous and automatic — the shadow experiment, gradual-rollout measurement and a threshold monitor running every five minutes. The review on this page is periodic and human. The detail of monitoring is under the policy's model monitoring heading.
Reviews are currently conducted in house. There is no independent third-party model audit yet; when one begins, it will appear on this page as a separate type.
Related pages
- Responsible AI Policy — the source of the review obligation
- Trust — compliance posture and the documents we can share
- Decision Governance — who publishes a rule and how it is rolled back