Response-time anomaly detection

3 min read · Notifications

ePulz.io does not just track whether a site is up or down. It also tracks response time and alerts you when a page suddenly slows down significantly compared to how it normally behaves. It is an early warning - slower responses often precede a full outage.

How it works

From the monitor's recent checks we build a baseline - the typical response time.
We compare each new response with the baseline using robust statistics (median and MAD - median absolute deviation). Unlike a simple average, it is not sensitive to a few random fluctuations.
If the response is significantly slower (it exceeds the modified z-score threshold) and at the same time rises by at least a fixed minimum value, we evaluate it as an anomaly and send an alert. The double condition prevents false alarms from minor fluctuations.

Where you will see it

The monitor detail shows the time of the last anomaly. The alert itself (e-mail, Telegram, webhook) additionally tells you how much slower the response was than usual (for example "2.4x slower"). The alert arrives through the same channels as outages.

Turning it on and off

Anomaly detection is optional - you turn it on in the monitor detail. It is available on the Pro and Business plans. Turn it off for monitors that naturally have very variable response times and would generate unnecessary alerts.

Tip: combine anomaly detection with a fixed slow-response threshold. Anomalies catch a sudden relative degradation, while a fixed threshold catches the absolute ceiling (for example "alert me when the response exceeds 2 seconds").

Was this guide helpful?