How often each service we track has actually stopped reporting normal service. These are our readings, taken every five minutes and kept, not figures any operator publishes about itself.
We poll each operator’s own status feed every five minutes and store the result. A service counts as down for a period when two or more consecutive readings show anything other than normal service. Readings where we could not reach a feed at all are excluded from the percentage rather than counted against the operator, because that failure may be ours.
Sorted least reliable first. Uptime covers the last seven days once we have that much history for a service, otherwise everything we have. Each service page carries its own incident log with start times and durations.