How we measure availability

At a set interval (from 1 to 60 minutes, chosen per website; at least 5 minutes on the Lookout plan) WWT sends an ordinary HTTP GET request to the website address with its own, clearly identified user agent. Redirects are followed, and every hop is checked for safety. The response time is the time from sending the request to receiving the response, measured from WWT's main server in Europe. The time limit is 10 seconds by default and can be changed.

A check fails when no connection could be made, the time limit ran out, or the server answered with a status code outside the accepted ones. By default 2xx, 3xx and 4xx responses are accepted and 5xx codes are failures: a server that answers "not found" is still running. The accepted ranges can be changed in the website settings.

Availability and the certificate are two separate checks: a website with an expired certificate that still answers is available, and the certificate problem raises its own alert.

After every failed check from WWT's server we ask two of our independent monitoring regions (Europe, North America, Asia, Oceania), rotated every hour, about the same website. If any region sees it working, the failure is treated as a local network problem and ignored. If the regions confirm the outage, or none of them can answer, the failure counts towards the threshold. A region whose own self-check has detected a connectivity problem takes no part in the decision.

An outage is confirmed when consecutive checks fail a set number of times in a row (3 by default, adjustable per website). Only then is an alert sent and an incident opened. Shorter runs of failures are not incidents and never alert.

The incident starts at the last successful check before the run: the website may have stopped working at any moment after it, so we count from that moment, against the measurement rather than in its favour. From the moment an outage is confirmed the website is checked every minute regardless of its interval, so the recovery is caught quickly.

An incident ends with the first successful check: from WWT's server, or confirmed from a region when our server still cannot see the website but the world can. The duration is the difference between the end and the start.

An incident also ends when we stop observing the website during an outage: after 7 days of continuous unavailability uptime monitoring switches itself off, the owner can switch it off by hand, and after a domain expires every check except the domain check is paused. Such an incident is marked "monitoring stopped" in the history; the real outage may have lasted longer, but WWT was no longer measuring it.

Availability for a period is the time the website was not in an incident, divided by the time WWT was actually observing it. An incident that started before the period or continues after it counts only for the part that falls inside the period.

Single failures below the threshold and failures contradicted by the regions do not reduce availability. Time when uptime monitoring was switched off is not observed: it counts neither as working nor as an outage. The measurement ends at the last check and starts again when monitoring is switched back on.

The average and median response times are calculated only from successful checks in the period. Failed checks neither lower nor raise this value.

Maintenance mode mutes notifications, not measurements: checks keep running, and incidents during maintenance are recorded and count towards availability.

Raw check results are kept for at least a year, currently the full history since the website was added. Incidents are kept for as long as the website is monitored, because they are the evidence.

WWT does not check the content of a page (no keyword checks), does not render it like a browser, and does not measure performance beyond the HTTP response time. Availability is measured from one point with confirmation from the regions, not continuously from every region. If you need a measurement from a specific location, tell us.