Return the materialized service dependency graph: every service, its current status, and the services it depends on. Call first to establish the causal structure before diagnosing. status in {healthy, erroring, unhealthy, restarting, hung, down}; monitoring='MISSING' flags a possible blind spot.

Localize the ROOT CAUSE deterministically: the DEEPEST failing service, the unhealthy one whose own dependencies are all healthy. Everything unhealthy above it is cascading fallout. Returns root cause(s), the causal chain per cascading symptom, blast radius, blind spots, and a page/no-page verdict, distinguishing a real outage from an observability blind spot (metrics missing but the service responds). Exact and repeatable, unlike per-investigation inference.

Transitive dependency closure of a service over DEPENDS_ON edges. direction='upstream': services that transitively depend on this one (its blast radius if it fails). direction='downstream': everything it relies on (trace toward a deeper root cause).

Detailed health snapshot for one service: status, container state/health, restarts, 5xx error rate, db pg_up, scrape health, and blind-spot flag.

List observability blind spots: services that are healthy but have no live Prometheus scrape target (lost visibility, NOT an outage - do not page).