This is the percentage threshold on CPU time spent performing actual, non-diagnostic, processing that, when exceeded, triggers a CPU-utilization alarm. Notably, this device's CPU is never truly idle; when not processing traffic it's continuously running diagnostic self-checks, so an admin reading this alarm knows it specifically reflects real traffic-processing load competing with those built-in diagnostics, not idle-versus-busy in the conventional sense. For example, a threshold of 90% firing during a traffic surge would flag that the node's CPU is now spending so much time on actual data processing that its background diagnostic checks are being crowded out.