The six components, live
A single green light tells you nothing useful: when it goes out, you still do not know whether your articles are being produced or whether it is only the website. Each component is therefore named and probed separately, with a real call rather than a box ticked in a configuration file.
Each component receives a real call, not a check that its configuration exists. The difference is everything: a key that has been revoked, has expired or has run out of quota passes the second check without trouble, so the page would show green for the whole outage, at the exact moment it is supposed to be useful. The calls consume no generation, and their result is cached for a minute so that a page opened during an incident does not make that incident worse.
What they do not say: we publish no uptime figure. Measuring one requires an external probe with history, which we do not have yet. A percentage worked out from memory would be exactly the number a customer would carry into their own tender, and it would commit them, not us. A component answering correctly does not mean it is fast either: these lights say “responds”, not “responds well”.
The register, and what changes after each incident
An incident here means a defect that reached production and was seen by someone. Each entry will say what the customer saw, what caused it, and above all the check that stops it coming back: “it is fixed” commits to nothing, “here is what now refuses it” commits to code.
Nothing has happened since this register was opened, on 6 August 2026.
That is not the same as “no incident ever”, and we are not going to let you believe it: the service existed before that date, it had outages, and they were fixed. This register begins on the day we decided to publish them. From here on, anything that affects a customer will be written down, with its duration when we measure it and “not measured” when we do not.
Something is wrong and it is not listed?
Write to us with the time, the article identifier if you have it, and what you saw. An incident nobody reports is an incident whose duration we will never know, and two minutes of your writing are worth more than our monitoring.
Monitoring from your side? Our health endpoint answers 503 when a component is down, and not only in the response body: an external probe left on its defaults will see it.