Unit 06.04: Recording who approved what, and when

After an incident, "was this approved?" needs an answer with a name and a timestamp on it.

What was shown, what was changed, what ran

Reviewer, decision, timing, edits, the evidence displayed, and whether the action executed.

Below is one approval record.

{
  "run_id": "run-8841",
  "action": "post credit 240.00 to ACC-1187",
  "presented_at": "2026-07-29T10:14:02Z",
  "reviewer": "operator-3",
  "decision": "approved",
  "decided_at": "2026-07-29T10:16:41Z",
  "seconds_on_screen": 159,
  "edited_fields": {
    "amount": {
      "from": "260.00",
      "to": "240.00"
    }
  },
  "evidence_shown": [
    "INV-1187",
    "payment log 4471",
    "payment log 4472"
  ],
  "executed": true,
  "idempotency_key": "run-8841:post-credit"
}

seconds_on_screen and edited_fields are what make fatigue measurable

evidence_shown is the field that separates two very different findings. An approval given on incomplete evidence is a system problem - the review card was missing something - while one given on complete evidence is a judgement call. Without the field they look identical.

seconds_on_screen and edited_fields are the same two numbers the previous unit tracks, recorded per approval rather than in aggregate. That is what lets you find the specific approvals that were reflexive.

The mistake this prevents

The mistake is recording the decision and not what was displayed. Six months later the review card has been redesigned twice, and there is no way to establish what the reviewer was actually looking at.

Takeaway

Record the reviewer, the timing, the edits, and the evidence that was shown. What was displayed is not reconstructable later, and it is what separates a system gap from a judgement call.

Mark complete