The 2 AM Argument
Imagine it is 2:00 AM. An engineer gets paged. The website is slow. They wake up the Incident Commander.
- Engineer: "I think this is a SEV1. It looks bad."
- Commander: "I don't know — only 5% of users are complaining. Maybe it's a SEV2? Let's not wake up the VP."
This debate wastes time. While you argue about the label, the fire burns. Every minute spent negotiating severity is a minute not spent fixing the problem.
Why Subjectivity Kills Response Time
Human beings under stress default to gut feelings. "It looks bad" is a feeling. "It feels serious" is a feeling. Feelings are inconsistent — the same incident will be called a SEV1 by an anxious junior engineer and a SEV3 by a battle-hardened senior who has seen worse. Neither answer is reliable.
The deeper problem is that severity determines who gets woken up, which runbooks get activated, and how much authority the Incident Commander has to make unilateral decisions. If severity is subjective, all of those downstream decisions become inconsistent too. You end up with some incidents over-escalated (burning out your on-call rotation with false alarms) and others under-escalated (a real SEV1 quietly destroying revenue while everyone assumes someone else is handling it).
The Need for Rigidity
A SEV Definition Matrix is a pre-agreed, written table that translates observable symptoms into a concrete severity number. It removes emotion from the decision. The on-call engineer does not decide the severity — the matrix decides it. The engineer just measures the inputs.
This is the same principle used in aviation and emergency medicine. An ER triage nurse does not decide whether a patient is critical based on how worried the patient looks. They measure blood pressure, oxygen saturation, and heart rate against a protocol. The protocol decides the priority.
Interviewers frequently ask: "How do you decide what severity to assign during an incident?" The weak answer is "I use my judgment." The strong answer is: "Severity should be determined by a pre-agreed matrix with objective, measurable criteria — things like percentage of users affected, error rate thresholds, or revenue impact per minute. Judgment is a fallback when the matrix doesn't cover an edge case, not the primary mechanism."
Key Point: Severity must be measurable, not subjective. The matrix is a decision tool, not a suggestion.