About this page

How a component is checked

A request runs from outside our own network, on a machine we do not own. It asks for a real page and checks that the answer contains something only that page carries, because a 200 proves something answered and not that the product did.

We do not publish how often that happens, because we cannot promise it. The checks are scheduled, the schedule is queued by somebody else, and it is sometimes late. Every row carries the time it was actually last measured, and the page tells you how long ago that was, so you can judge it rather than take our word for the interval.

Each bar shows the last 90 days, one cell per day, coloured by the worst state we measured that day. A day with one hour of trouble and twenty-three of health is shown as trouble. Days before we started recording are blank rather than green.

What the states mean

Operational is a check that reached it and got what it should. Degraded answered, more slowly or less reliably than it should. Unknown means we tried and could not get an answer, which is not the same as working and not the same as broken. Not measured means nothing watches it yet, and we list those rows so you can see the gap instead of assuming coverage.

Why there is no uptime percentage

Every number of that kind we could publish would be derived from our own account of our own incidents, which is the proposer checking its own work. We show you what each check found and when instead.

Why there is no explanation of an incident

The records on the history page are machine records. When a check changes its answer we write down the component, the time and the new state. Nobody has written an explanation of why, because there is no operator typing updates here yet. When that changes, the written account will appear alongside these timings rather than replacing them.

Notifications

There is no email sign-up. A feed collects no personal data, so there is no consent record to keep, no revocation path to build and nothing to erase. Changes are published at history.atom.

Where your data lives

The database holding your company data is in Zurich, Switzerland. Requests to the models we use are processed by Anthropic and OpenAI, outside Switzerland and outside the EU. The evidence archive is pinned to the EU.

What this page cannot tell you

If the status page is unreachable, look at our status repository. It is not served from our own infrastructure, so a problem with ours does not take it down too.