Unified telemetry on HUMΛN HumanOS
Canon: kb/167_unified_telemetry_observability.md
API standards: kb/101_api_design_standards.md
Why one plane
Agents, workflows, connectors, and the HTTP API should emit one correlatable stream — not parallel logging silos. The unified plane feeds Command Plane Executions, Events, Incidents, Partner health, and optional webhooks.
Envelope families (short)
| Namespace | When |
|---|---|
telemetry.run.* |
Gateway / async / connector host lifecycle |
telemetry.author.* |
ctx.telemetry.emit from your code |
telemetry.api.* |
Classified client errors (RFC 7807 path) |
telemetry.platform.* |
Operator-only / platform_internal |
Where rows land
events— bus + fan-out towebhook_deliveriesandagent_triggerasync_executions/cp_executions— run mirrors for the CP consolecp_incidents— incidents, remediation, tracker links
Use execution_id, trace_id, and stable HumanError code fields so subscribers join API errors to runs without leaking stacks to tenants.
Scale
Fan-out is enqueue-only; delivery workers retry with backoff. Prefer scoped event patterns over subscribing the entire telemetry.* firehose until you need it.