Interpreting platform status

Agents that read the platform’s status endpoints often misread normal idle states as failures. This page lists the interpretation rules. When your conclusion is “the platform is broken”, check this table before reporting.

Rules

You see	It means	It is NOT
orchestrator.present: false / alive: false	Normal idle. Production has no permanently-running orchestrator daemon; generation runs on a nightly timer	An outage. Alarm only when present: true and alive: false (a heartbeat went stale)
velocity: 0, in_flight: 0	No run is active right now	A stuck queue
contracts[].remaining > 0	Generation candidates the nightly batch will work through under its daily cap	A backlog of failures, or an embeddings queue
A quiet period during the day	The nightly run finished; the next one starts on the timer	A dead scheduler

Where to look for the real execution funnel

• The cycles domain (see list_domains) shows the actual run funnel: what was dispatched, what completed, what failed.
• Recent failures live in the error surface: GET /api/v1/errors (see How to read errors).
• Every run is recorded as pipeline_run with per-step run_step rows — a run that failed is visible there, not inferred from idle status.

The golden interpretation rule

Meaning lives in payload values. When a status field carries a note alongside a number, the note is authoritative context — quote it in your report instead of inventing an explanation for the number.