wiki/decisions/ADR-2026-06-08-n8n-backend-control-plane.md
ADR: Reinstate n8n as Backend Workflow Control Plane
Date: 2026-06-08
Status
Accepted
Decision
n8n is reinstated as the backend workflow/control plane for CurioQuest Book Factory operations.
The old 12-agent/n8n architecture remains rejected. What is back is n8n as the operational backend surface: workflow orchestration, visibility, manual control, retries, and inspection. Code services remain typed capability providers that n8n can call for catalog lookup, generation, persistence, artifact handling, validation, and helper APIs.
Context
The founder rejected continuing the backend as opaque code orchestration because it reduces observability and direct operational control. The pilot needs a system the founder can inspect, intervene in, and reason about while books are being produced.
Recent n8n intake work already established a useful boundary:
intake form -> /api/intake -> n8n webhook -> database/workflow
That direction should now be treated as architecture, not an experiment.
Rationale
- n8n makes workflow state visible to the founder/operator.
- n8n supports manual inspection and control during the concierge pilot.
- n8n reduces the risk of an invisible backend pipeline failing without an operator-friendly recovery path.
- Code remains appropriate for deterministic, typed, testable units; n8n should orchestrate those units instead of replacing their internal correctness.
Alternatives Considered
| Option | Why rejected |
|---|---|
| Keep opaque code-first backend orchestration | Founder explicitly rejected it as too hard to observe and control. |
| Restore the old 12-agent/n8n architecture wholesale | The old business model and 12-agent architecture remain superseded; only n8n as workflow/control plane is reinstated. |
| Mirror all n8n documentation locally | Too likely to go stale. Keep online docs as source of truth and store only CurioQuest-specific n8n decisions/patterns locally. |
Consequences
- Docs must no longer prohibit n8n.
- Backend workflow work should start from
docs/n8n/README.md. - n8n workflows need schema/version discipline so workflow convenience does not become hidden contract drift.
- Code services should expose stable, test-covered boundaries for n8n rather than silently accumulating orchestration logic.
- Future backend verification must include both code-level tests and n8n workflow smoke evidence when the workflow is in scope.