050673 status

Each endpoint is called from outside the cluster once a minute, over public DNS and real TLS, exactly as your own client would reach it. We send no API key, so a rejection is the healthy answer — it proves DNS, the edge and authentication all worked. Latency is the whole round trip including handshake, so it will read higher than your own once your connection is warm. The limitation worth knowing: these probes run on our own infrastructure. If everything of ours were down at once, nothing would run and no outage would be recorded here. A page that cannot report its own worst case should say so.

Upstream health is the gateway’s own view of the chain nodes behind these endpoints — we report it rather than measure it. A degraded upstream usually means slower responses rather than errors, because traffic is moved off it automatically.