Dependency records

Observed state

From the five most recent observations of the listed endpoint: responding when all five received the expected response, degraded when any did not, and “not observed recently” when the newest observation is older than fifteen minutes. A state describes the endpoint, not the vendor’s whole service estate.

Last observation

The most recent completed check. Freshness, not health.

Region labels

The region labels scheduled for the dependency. A label names the worker that ran the probe, and RELIASTRA operates one observation point today, so two labels are not two independent origins and nothing here is corroboration. Confirmation across genuinely separate sites is a property of deployments that have more than one.

Independence

Every figure originates from a RELIASTRA probe. The status text on vendor status pages is not ingested - where one is listed as an observed endpoint, only its HTTP behaviour is measured.

Scope

Public endpoints only. Customer endpoints and credentials never appear here. An empty incident list is an absence of published records, not proof that no outage occurred.

Method: How RELIASTRA measures vendor reliability. Topic hub: AI infrastructure status & reliability.