How Outages Watch works

What is read

Each service on the board publishes a public status page with a machine-readable API. Once a minute, a scheduled job on our server reads 2 of them, in rotation, so every page is read again about every 9 minutes. It requests the summary of the page: the overall indicator, the sentence the provider attaches to it, the open incidents and the list of components. Two providers, Cloudflare and Twilio, publish several hundred components; for those only the overall indicator and the open incidents are read.

The answer is validated before it is used. A response that is not the expected format counts as a failed read, exactly like a timeout. Nothing is inferred from a failed read.

What your browser loads

The result of each run is stored as one small file. Your browser loads that file from this site and asks again once a minute while the tab is visible. It does not contact the providers. The page you receive first contains no status at all, only the text; the states are filled in by that file, so an old cached page can never show an old state as current.

To limit writes, the file is rewritten when something changes and otherwise every 10 minutes. The "last confirmed" time on the board is the time of that file.

When there is no answer

After 3 failed reads in a row (about 27 minutes of silence), a service changes to No data: a hatched grey lamp with those words. After 5 failed reads the collector raises an alarm so the source can be checked. If the collector itself stops and the file becomes older than 25 minutes, every service on the board is shown as No data and a notice says since when.

The rule behind this is simple: silence is not health. A service is shown as operational only when its official page has just said so.

History and the 90-day strip

About every 18 minutes per service, and at once when an incident opens or changes, the collector reads the provider's official incident list and merges it with what it already holds. Incidents stay in the record after they drop off the provider's own recent list.

The strip colors each UTC day by the most severe incident declared for it. The record starts at the oldest incident we hold for that service; earlier days are grey, meaning "no record". The time figures on each service page add up the periods covered by declared incidents, counting overlaps once.

What this cannot tell you

• Undeclared problems. If a provider does not post an incident, this board shows no incident. Status pages are updated by people and trail the start of a problem.
• Whether it works for you. Regional failures, account limits and local network problems do not appear on a status page.
• Latency or uptime percentages. Those need independent measurement, which this site does not do.
• Any website you type. Only the listed services are collected. The site makes no requests to addresses supplied by visitors.