Service status
Snapshot generated 2026-08-16 21:55 UTC · republished with each site data build.
The last patrol verdict
Green — every check on the last live patrol pass came back passing.
Last live patrol: 2026-08-16 18:42 UTC.
An automated patrol probes the public service from outside on a fixed cadence — reachability, TLS, the answer endpoint, the MCP endpoint, content freshness — and records one verdict: green when every check passes, yellow when a non-critical check fails, red when a critical check fails. This page renders that record as committed; it is a snapshot, not a live probe. The machine-readable envelope behind it is served at /ops_health.json, and an external monitor reads the same file on its own clock, from outside our infrastructure — a stalled build is itself an alarm someone hears.
The operating record, at this snapshot
- Live patrol — verdict Green, recorded 2026-08-16 18:42 UTC.
- Static export audit — last completed 2026-08-16 17:40 UTC.
- Scheduled pipelines — liveness not recorded in this snapshot.
The machine door
One MCP endpoint carries the whole desk — reads are free and keyless; the Answer Engine rides a free self-issued key. The patrol verdict above covers this door: an orient call on the endpoint must answer, and that check is graded critical.
POST https://api.signalbureau.ai/api/mcp # MCP (JSON-RPC 2.0, streamable-http)
- Discovery manifest: /.well-known/mcp.json
- Integration map — tools, tariff, the async ask pattern: signalbureau.ai/machine/
- This page, machine-readable: /ops_health.json
What we commit to, honestly
The service is continuously operated and monitored by automated patrols. There is no contractual SLA on the public service today, and we will not publish an uptime number we have not measured and committed to. We expect formal service commitments to arrive when metered accounts reach general availability; service commitments for contracted private desks are set in the contract. The full posture — data classes, retention, isolation, availability — is stated on the trust page.
Seeing a failure this page does not show? Tell the desk — machine callers can file it in one call with the send_feedback tool on the MCP endpoint. A complaint that names a real defect becomes our work order.