Every endpoint, and how it is actually doing

Success rate is the share of requests that returned data, over the last 30 days, excluding requests that failed because of the parameters you sent, and requests refused with a 429 because too many were in flight. A 429 costs nothing and says when to try again. A 404 for a profile that genuinely does not exist is a success: we answered the question correctly.

Latency is measured from when we start answering a request, after its key is checked, to when its answer is recorded, and excludes cache hits — which would flatter it considerably.

Whose calls. Real calls only: the no-key demo and our test accounts are left out. A call our own account makes for its own work is a real call, and counts. An endpoint shows a number once it has 50 calls in the last 30 days.

Operational means we can serve the endpoint now, and its success rate over the last 24 hours is at least 95% where it had 50 calls in that time. The state is about now; the 30 days’ success rate beside it is the record, and keeps a bad day for a month. Where we can, an endpoint is served by more than one source, so a problem at one of them is invisible to you.

Degraded means we can serve it, but its success rate over the last 24 hours is under 95%, or every source for it has just failed three times in a row and is paused for a minute, or our nightly check of it failed, or at least three calls to it failed in the last half hour. Unavailable means we cannot serve that endpoint at all — and calls to it are not charged.