Health observations

The Gateway times every forwarded attempt. Reported latency is the mean of successful attempts over a recent rolling window, and health status uses separate thresholds for entering an unhealthy state and for returning to a healthy one. Observations decay to unknown once an Endpoint stops receiving traffic. Upstream 425 and 429 responses are excluded from health scoring.

Health status and latency are observational. Circuit controls runtime fault admission, while configured weights control load balancing. The routing behavior is described in Routing and reliability.

The Endpoint health check in the dashboard is a manual, on-demand probe. The Gateway does not probe idle Endpoints on a schedule, so an Endpoint that receives no traffic carries no fresh observation.