Take the five-minute drill: how fast can you reconstruct one AI approval?

Here is a test worth running before an auditor runs it for you.

Pick one production AI system your organisation approved — any one. Now, from records alone, rebuild the approval: who decided, on what evidence, under which version of the rules, and with what conditions attached. Time it. Five minutes is the standard we work to.

When we talk to governance and audit practitioners, the honest answer is usually that the digging starts immediately: chat threads, meeting minutes, a decision someone remembers being in the room for. That digging is archaeology, and it is what an auditor finds when the machinery has no memory.

So, a genuine question for the practitioners here: how long does the reconstruction take in your organisation — and where does it break down? Is it the decision itself that was never written down, the evidence that was reviewed but not kept, or the conditions that were agreed verbally and never attached to the record? And a related one for the 2 a.m. case: if a customer-facing system had to be halted tonight, is it written down anywhere who holds that authority?

The discipline we work from has two rules. No decision is communicated before its record exists. And the drill itself, run monthly and unannounced — pick one approval, rebuild it from the evidence in five minutes — with the pass rate treated as the program's core honesty metric: the number a board should ask for, because it measures whether the records are real or decorative.

We would rather learn from your numbers than talk about ours. Run the drill, tell us the time, and tell us what was missing.

—

For teams who want the shapes rather than inventing them: we publish AI Governance in Practice — Part 2, 42 templates, checklists and worksheets for the decision-and-evidence half of the discipline — a one-page decision-rights map and CAIO charter, the decision record, and the design sheet for an append-only evidence store. They are documentation templates and design guidance: no software ships with the pack, the evidence store is a design your team implements, and no evidence is generated automatically — the records exist because your process writes them, in the shapes the templates specify. The pack also carries the source book's own EU AI Act / NIST AI RMF / ISO/IEC 42001 crosswalks, carried exactly as printed; they are not legal advice and certify no compliance with any regulation.

https://aisoltrics.gumroad.com/l/ai-governance-audit-templates?utm_source=devto&utm_medium=post&utm_campaign=ra-governance-202611&utm_content=day1-governance-part2-devto

— AISoltrics Technologies Inc.