The status beside each component is a live check made when this page was rendered, from our servers, unless an incident above declares otherwise — a written incident always outranks an automated check. A component we do not probe is marked Declared, and a check that could not run reports Status unknown, never “operational”.

One limit worth stating plainly: the website row is partly circular. This page is served by the very deployment it reports on, so a total outage means the page is not served at all and can never be the thing that tells you. It does catch a broken route or a bad build while the platform is up. The rows that are checked — the backend, accounts and the release feed — are measured across the network and carry no such circularity. Desktop review runs is neither: it is declared, because nothing of ours sits in that path to measure.

Uptime is derived from the incident list on this page, not from a separate monitor: a major outage removes its full duration, a partial outage removes half of it (partial_outage = 0.5, major_outage = 1), and announced maintenance removes none. Days before we began keeping records are drawn as blanks and excluded from the percentage rather than counted as good days.

Record keeping began 2026-08-09. The same data is available as JSON at /api/status, which is what the desktop app reads.