Platform status and health
The health endpoints
Section titled “The health endpoints”curl -fsS https://entherium.duckdns.org:8443/healthzReturns {"status": "ok", "service": …, "version": …} — the version
field doubles as “which build is live”. /health is an alias.
| Signal | Meaning |
|---|---|
200 on /healthz | Platform up. Everything else on this page is about interpretation, not availability |
502 / connection refused | The platform instance is not serving — down or intentionally paused (maintenance windows exist; a paused instance is not an incident) |
200 but idle metrics | Normal between nightly runs — see below |
Live status
Section titled “Live status”/api/status.json…GET /api/status.json (no token) renders the same data in the Statuspage v2
summary.json shape — status.indicator (none / minor / major /
critical), components[] for the three public surfaces (API, MCP, Developer
docs) and page.updated_at — so any status widget or agent that already reads
that format works unchanged. It answers 503 when the database is down and
sends Access-Control-Allow-Origin: *.
curl -sS https://entherium.duckdns.org:8443/api/status.json | jq '.status, [.components[] | {name, status}]'Freshness of this documentation
Section titled “Freshness of this documentation”Every page shows the API contract version and the agent capability version it
was built against (the badges under the title). The build itself is described
by versions.json (version, builtAt) and
manifest.json (a content hash per page); the Developer docs component above carries the portal version the API host currently sees.
Idle is the default state
Section titled “Idle is the default state”Production runs on a nightly batch; between runs there is no orchestrator
process by design. The definitive interpretation table lives at
Interpreting platform status — read it
before reporting an outage based on orchestrator.present: false,
velocity: 0, or a quiet afternoon.
Deeper surfaces (authenticated)
Section titled “Deeper surfaces (authenticated)”| Surface | What it shows |
|---|---|
GET /api/v1/ops/overview | The live operational snapshot: orchestrator state, per-contract backlogs, alerts |
GET /api/v1/errors | The durable error sink (how to read it) |
The cycles domain | The actual execution funnel of the nightly runs |
Run records (pipeline_run / run_step) | Ground truth for any specific execution |
If you operate through MCP
Section titled “If you operate through MCP”A platform deploy kills live MCP sessions without restoring them — an
empty tools list right after a version bump is reconnect-time, not an
outage (details). After reconnecting,
start_here tells you if the capability surface changed while you were
away.