When your phone buzzes at 02:14 in the morning and the alert says "CPU at 87%", you don't actually want to know the number. You want to know: is something broken, was it the deploy at 02:11, and what do I do about it?
Descriptive monitoring (Datadog, New Relic, Grafana out of the box) is built around the first question. It's great for capacity planning. It's noisy for incident response.
Causal monitoring takes a different tack: correlate the metric anomaly with deployments, log signals, and per-process attribution, then return a single sentence with confidence.
That's the wedge cpum.ai is built around. The 87% is still in the data — it just isn't the answer.