Rounds
Each scheduled round sends the check from six probes, spread over the fleet. A scheduled round never declares an outage on its own. If any probe reports a failure, confirmation rounds start about eight seconds later and repeat every twenty seconds, this time from every healthy probe in the fleet. Confirmation rounds cost nothing.
The down rule
A monitor is confirmed down when, in two confirmation rounds in a row:
- at least two thirds of the healthy probes fail, for example 8 of 12;
- the failing probes are on at least three continents;
- and on at least three different providers.
At most four confirmation attempts are made. If fewer than eight probes are healthy, or they cover fewer than four continents, no down verdict is given at all; the fleet would rather stay silent than guess. The monitor page shows the rule it is under as "down = N of M probes". The probe list shows the fleet.
Continents can be excluded from voting on a monitor, for example a site that is only served in Europe, as long as at least three probes on two providers remain.
Other verdicts
- Partial: the target is unreachable from some continents only. At least two probes fail, at least half of the probes in the affected continents fail, and fewer than a third fail elsewhere. Partial is shown in the panel and alerts only if the monitor's Alert on partial outages is on.
- Blocked: a firewall or Cloudflare answered the probe with a challenge or a block (a 403, 429 or 503 with a challenge page, or Cloudflare errors 1005, 1006, 1010 and 1020). This is a warning, not an outage, because your visitors may be fine. While a monitor stays blocked, repeated blocks do not start new confirmation rounds. Channels can alert on it ("Also alert when a WAF blocks our probes", on by default). The Cloudflare check tells you in advance whether your WAF treats the probes this way; our probe IPs are public so you can allow them.
- Slow: the median response over the threshold for two rounds, with at least three samples. A warning; see Monitor types.
- Warnings such as a certificate about to expire or a missing SMTP banner count as up.
What counts as a failed check
A status code you did not accept, an error page when detection is on, a keyword rule that fails, too many redirects, a timeout, or no connection at all. An expired certificate is a failure.
Incidents
An incident opens when a monitor is confirmed down, one at a time per monitor, and closes when the monitor moves away from down: a confirmed recovery, a maintenance window starting, or the monitor being paused. The Incidents page opens on the open ones, with the full history under Log. An open incident can be put on hold while someone works on it: the "still down" reminders pause and are not counted against the limit of 15, monitoring and the recovery alert go on as usual, and the row shows who put it on hold and when. Each incident records when it started, when it was confirmed, from how many locations, and how long it lasted.
A round where at least two probes on two providers failed without a confirmed outage is kept as an unconfirmed failure: an orange mark on the monitor's timeline and a counter, with no effect on uptime or alerts. It is the trace of a flaky route or a server that stumbles for a second.
Uptime
Uptime is 1 − time down ÷ time with a known state. Only confirmed down time counts as down; partial and blocked do not. Time while the monitor was paused, in maintenance, or had no result is left out of both sides. The monitor list shows 24 hours; the monitor page shows 24 hours, 7 days and 30 days, with hourly bars for the day and daily bars for the month.
Last change: 31 October 2026.