How this page works
A check is one HTTP request from one of two monitoring locations, every 30 seconds. A single failed check is normal on the open internet, so we don’t report it. It takes a pattern.
Degraded means one location has missed more than 5% of its checks over the past hour and is still missing them right now, or the service is answering but has been consistently slow for the last ten minutes. Slow to alarm, quick to forgive: an hour of evidence to light up, cleared within ten minutes of the last failed check. Down is immediate. If nothing anywhere can reach it, that isn’t noise.
The bars are the last 30 days, one per day, and follow the same idea. Amber where a location spent more than five minutes of that day missing checks. Red only where the site was unreachable from both locations at the same time for more than a minute. Two locations having trouble at different moments isn’t an outage, because something could still reach the site throughout. Grey is a day with no monitoring data, which is a gap in our records rather than a bad day. Hover any bar for what monitoring saw.
The uptime numbers are the share of checks that succeeded over the last 24 hours, 7 days and 30 days, taken from whichever location had the best view. A flaky path between us and one monitoring box shouldn’t count against a service that was up. They’re measured over the checks that actually ran, so days with no monitoring data neither help nor hurt the figure. We colour them green at 99.9% and above, amber from 99%, and red below that.
Freshness. The page reloads every 60 seconds while the tab is open. The server re-queries monitoring at most once every 30 seconds, because asking faster than the probes run can’t produce new information. Monitoring runs on self-hosted Prometheus and blackbox_exporter, the same stack we deploy for open source clients.
Want to know what we monitor and how quickly we respond? That’s on Security & Uptime.