The SEC examiner requests the AI decision record.
The record is already signed.

The scenario · 09:14 AM, Monday

Equity-routing flag on block-trade #04-7421.
Six questions. One ledger.

1. 1.What model produced this decision — and which version? The SDK tags the model identifier, the commit, and the routing-policy version at inference time — captured inline, not reconstructed from a CI log five months later.
2. 2.What score triggered the flag? Three evaluators ran inline: a venue-impact rubric, a wash-trade-overlap gauge, and a regulatory-threshold ceiling. Each evaluator signed its result to the same row.
3. 3.Is the threshold policy reproducible? The escalation threshold, the throw rate, and the policy hash are all on the same record as the model output — three lookups, not three vendor portals.
4. 4.Has the model behaved the same way all quarter — or is there drift? Drift ledger pulls the same evaluation key against the last 90 days — calibration curve walks the examiner back in a single view, and any incident on the way is on the same chain.
5. 5.Who has access — and is the access trace in the row? Scope is set at write time and recorded with the row. `compliance`, `risk`, `exam-read-only` are distinct call paths, and every retrieval leaves its own audit crumb.
6. 6.Will the record hold up under subpoena? SHA-256 signatures anchored to a key not on the production write path; tampering is detected at fetch time, not at audit time.