The four questions every governed AI decision can answer

Here is the test for whether your agency can stand behind its AI.

For any AI-influenced decision, can you answer four questions without reconstructing the story afterward?

1. What happened

Is there a record of what the AI actually did, captured as it happened — not summarized later from memory or pieced together from logs that were never meant to be evidence?

2. When it happened

Can you prove the timing, with a timestamp no one can quietly change? "Around then" is not an answer an auditor accepts.

3. Who set the rule

Can you name the role that authorized it, with permissions bound to that authority? Accountability requires a named owner, not a diffuse "the system did it."

4. Why it was valid

Was a condition checked before it ran, and was the proof kept? A governed decision carries its own justification.

---

Answer all four, and the decision is governed. Miss one, and it is exposed.

Most agencies can answer one or two today. The distance to four is the work — and it is the same distance an appeals board, a records request, or a reporter will measure when they come asking.

These four questions are the Trust Rails test, translated from AI Native by Design into the public sector. We hold the field to them, and we hold ourselves to them.