JUN 14, 2026
A fixed latency threshold of "alert above 500ms" treats a database proxy and a static asset endpoint identically, even though their normal operating ranges differ by an order of magnitude. We walk through how baseline-relative alerting changes the false-positive rate in practice.
MAY 02, 2026
Our original scheduler used a naive cron-per-check model that stopped scaling around fifty thousand concurrent checks. Here's the redesign that got us to several million.
MAR 19, 2026
Yes, the monitoring company's status page broke. Here's exactly what happened, why our architecture didn't catch it sooner, and what changed afterward.