Sensitivity & dismissal

Not every spend movement is a problem. The sensitivity selector decides how aggressively CloudQuell flags anomalies, and dismissal lets you clear the ones you’ve confirmed are expected — so the feed stays signal, not noise.

Note

Sensitivity tuning and dismissal are paid features. On the Free plan they’re locked behind an upgrade prompt. See Plans & billing.

Tuning sensitivity

The dashboard’s anomaly feed has one detection control: a sensitivity selector with three levels. It scales the thresholds every detector uses — how large a percentage move and how big a dollar change have to be before a service is flagged — and your choice is remembered in your browser.

• Low — fewer alerts — raises the thresholds, so only large, high-confidence spikes surface.
• Balanced (default) — the standard thresholds: roughly 30% month-over-month, 40% week-over-week, and 30% for the live rolling-30-day window, each paired with a dollar floor.
• High — flag earlier — lowers the thresholds, catching smaller moves sooner (more, smaller anomalies).

Changing the selector re-runs detection — it is not a display filter. To hide already-flagged rows without changing what’s detected, use the window chips and the minimum-severity filter on the feed instead.

Via the MCP tool

Through the CloudQuell MCP server, the list_anomalies tool exposes the raw detection knobs directly rather than the three-level selector: a minimum percentage increase (the tool’s own default is 25%), a minimum cost dollar floor, and a service filter to narrow to one area.

Dismissing an anomaly

When a spike is understood and expected — a known backfill, a planned migration, a one-off load test — you can dismiss it. Dismissal suppresses that anomaly so it no longer surfaces in the feed. A dismissal is keyed by the service, the period it occurred in, and the detection window (Live, Weekly, or Monthly) — all three identify the exact anomaly, so the same value is needed to restore it — and you can optionally record a reason for the record.

Dismissing with an agent

Through the CloudQuell MCP server, dismissal is a plain-language request:

“That RDS spike is a known one-off backfill — dismiss it.”

Dismissal is a write action. It requires an admin:write token and an explicit confirmation on the call, so an anomaly can never be suppressed without an intentional, confirmed step.

Tip

Dismiss understood anomalies, don’t just raise the threshold past them — a reason on the dismissal keeps the audit trail, while a blunt threshold change can hide unrelated spikes too.

Next steps

• Anomaly detection — how anomalies are surfaced in the first place.
• Investigating anomalies — confirm a spike’s cause before you dismiss it.