How detection works

We check, we do not wait to be told

Most outage sites depend on visitors reporting problems, so a service nobody has complained about yet looks fine. NetOutages checks every service itself, so a brand-new site has a status from its first minute, with no reports at all.

Every service we list is requested once a minute. The websites you add yourself are checked every 1, 5 or 10 minutes, as you choose. Each check measures the lookup of the domain name, the connection, the secure handshake, the time to the first byte of the answer and the total time.

What counts as down

A check fails when the site does not answer in time, the domain does not resolve, the connection is refused or the secure handshake fails, when it answers with a server error (HTTP 500 and above), when it answers with a status other than the expected one, or, if you asked for it, when a word you require is missing from the page.

A site that answers, but takes 2.5 seconds or more, is shown as slow rather than down.

Sites sometimes redirect before they show a page. We follow up to 10 redirects, and every step is checked for safety again.

How we avoid false alarms

• A failed check is repeated at once. A single hiccup is not recorded as a failure.
• When something fails we also request a few well-known sites. If none of them answers either, the problem is our own connection, and the whole round of checks is thrown away.
• A service is only called down after three failed checks in a row, and only called back up after two successful ones, so a flapping site does not open and close an incident every minute.

When a site blocks us

Some sites refuse automated visitors with a challenge page or an error such as HTTP 403 or 429. That says nothing about whether the site works for people, so we never count it as an outage. Such a site is marked “limited”, we say so on its page, and for those sites user reports are the only live signal. We do not pretend to be a browser and we do not solve challenges.

A few sites do the opposite: they hold automated requests open until they time out. From one place that looks exactly like an outage, so for the sites we know do this, timeouts alone are not enough. We wait for users to report problems, or for a different kind of failure, before calling an outage, and the page says that we are waiting.

What users add

On every service page you can report a problem. We store the kind of problem, the time, an approximate location and a one-way code that stops the same person reporting the same service more than once every 10 minutes. Not your address.

Reports never change a service's status. They tell you when other people disagree with it: when at least 10 people report within 15 minutes, and at least three times the usual number, the page says users are reporting problems even if our checks pass.

Locations are worked out from the visitor's address and are approximate: a VPN or a mobile network can put someone in another city. A city is shown only when several people reported from it.

What we cannot tell you

• Our checks come from one place. A problem that only affects some countries or some networks may not show up, or may show up as an outage for everyone.
• A site can answer the page we request and still be broken somewhere else, such as a login, a payment or an app. Reports help there.
• The threshold for “slow” is the same for every site, and we do not yet learn what is normal for each one.