Method

How we measure. Disclosed.

Measured, not claimed: Status comes from automatic checks: reachability of the pages, our health endpoint /api/health with database and sign-in, and external monitors where needed. Nobody sets it to "operational" by hand.
No alarm from a single error: A check only counts as disrupted after three failures in a row and as healthy again only after two successes in a row. Single dropouts trigger no public disruption.
From check to service: If a service's reachability fails or two critical checks fail at the same time, that is a major outage. A critical check alone is a partial outage, less important checks or slow responses mean degraded performance.
Availability: Available is the time a service was operational or only degraded, divided by the measured time. Partial and major outages count as downtime. Scheduled maintenance is shown separately and not counted as an outage.
No invented history: Before the first check there are no values. If measurements are missing, for example because the measuring service wasn't running, that time counts neither as available nor as downtime; in the history it appears gray.
Periods: 24 hours from the running history, 7, 30 and 90 days, one year and the time since measurement began from the daily journal (calendar days in UTC, today included). The daily journal is stored permanently, individual checks for 90 days.
Manual interventions: People write incidents and maintenance and can correct a measured status with a reason, for example in case of a false alarm. Every such change is logged.