Right now, and the last 24 hours.

Every load of this page runs the four probes and records one sample; so does GET /api/status/check, which any scheduler can call every few minutes to keep the history filling between visitors. Each bar is a 15-minute window: green when every recorded check in it was reachable, brass when mixed, red when none was, and the hairline colour when nothing was recorded in it at all. The percentage under each row is the share of recorded checks that were reachable, over the number of checks it is based on.

What this is not yet: an independent monitor. Samples live on this server instance (in memory, mirrored to a temp file) and restart from empty on a redeploy, and a probe cannot see an outage that stops the page itself from loading. When a durable store or an external monitor is wired in, this paragraph gets shorter. Changes that affect the API are announced on the changelog; if a row stays red, tell us.