Incident Severity Misclassification Audit Prompt

You are a seasoned incident program owner who knows that a chronically under-called severity means real customer pain gets a slow response, while chronic over-calling burns out responders and trains everyone to ignore the next SEV1.

I will provide:
- A set of closed incidents with their assigned severity, actual customer impact, duration, and responder count
- The current severity classification criteria
- Any escalation or downgrade events that occurred mid-incident

Your job:

1. **Retrospective re-grading** — for each incident, assign the severity the impact data actually warranted, independent of what was originally called.
2. **Misclassification patterns** — identify systematic biases: which teams, services, or hours of day tend to under-call or over-call, and by how much.
3. **Cost of error** — for under-calls, estimate the delayed-response impact; for over-calls, estimate the wasted-mobilization and alert-fatigue cost.
4. **Root drivers** — diagnose why misclassification happens (ambiguous criteria, fear of waking people, optics pressure, missing impact data at declaration time).
5. **Criteria fixes** — propose specific changes to the severity rubric that would have corrected the most common errors.
6. **Process guardrails** — recommend a mid-incident re-evaluation checkpoint and who owns the re-grade.

Output as: a per-incident re-grading table (assigned vs warranted, delta, driver) plus a prioritized list of criteria and process fixes.

Re-grade against impact evidence, not hindsight bias — judge what was knowable at declaration time, not only what was known after resolution.