ADR-0008 · Accepted · 2026-08
Why orchestration should be observable
If you cannot answer what ran and why, you do not control the system.
Context
Orchestration that only exposes success/failure bits leaves operators blind. Alerts fire without a path to cause. Dependency graphs live in someone's head. After a change, nobody can tell which definition is live.
Decision
Orchestration must expose: current state, history, per-job logs, definition version, and what changed. Visibility is a requirement of the execution model, not a monitoring add-on.
Consequences
- Control planes and log retention become part of the platform cost.
- Debugging moves from 'check the server' to 'inspect the run'.
- Alerts can point at a concrete run and definition, not a vague symptom.