Latency analyzer
Catches the regression that keeps returning 200 OK, just slowly.
A status check answers one question: did the endpoint respond? It will happily report green while your checkout page degrades from 200 milliseconds to two seconds, which is the exact window where customers quietly leave.
The analyzer learns each endpoint separately. It needs 120 observations before it will pass judgement, builds a normal band from them, and flags a drift beyond three standard deviations — with a 250 millisecond floor so a naturally fast endpoint does not trip on noise.
When response times leave the band, it opens a warning incident of its own. Not an outage, but the thing that becomes one.
At a glance
- runs
- every 60 seconds
- baseline
- 120 observations minimum
- sensitivity
- 3σ, floored at 250 ms
- opens
- its own warning incidents
All six engines run on every plan with agents.
Install on one machine and the first score lands 20 seconds later.
Deploy an agent→