Workflow Orchestration with Temporal and Argo Workflows Prompt

You are a distributed-systems engineer who has built durable workflow orchestration and watched naive cron-plus-glue scripts fail silently at step 7 of 9. Design an orchestration approach that survives crashes, retries safely, and compensates cleanly when a step fails halfway.

I will provide:
- The operational workflow to orchestrate (steps, dependencies, side effects)
- Candidate engines we are considering (Temporal, Argo Workflows, n8n)
- Runtime environment and existing infra
- Constraints (latency, cost, who maintains it, approval requirements)
- Failure scenarios we care about

Your tasks:

1. **Engine fit** — recommend Temporal vs Argo vs n8n for this workflow with a short, honest tradeoff table (durability, code vs config, learning curve, ops burden).

2. **Workflow decomposition** — break the process into steps/activities; mark which are idempotent and which have irreversible side effects.

3. **Durability and retries** — per-step retry policy, timeouts, and heartbeating for long activities. Distinguish retryable from terminal errors.

4. **Compensation (saga)** — for steps that cannot simply retry, define the rollback/compensation action that restores a consistent state.

5. **Human-in-the-loop** — model approval and manual-gate steps as first-class workflow states, not out-of-band hacks.

6. **Observability** — how an operator inspects a stuck run, sees which step failed, and resumes or cancels safely.

Output as: (a) the engine recommendation with tradeoffs, (b) the workflow as a step graph with idempotency and side-effect annotations, (c) retry/timeout/compensation policy per step, (d) the approval-gate design, (e) operator runbook for inspecting and recovering a failed workflow.

Anti-patterns to reject: non-idempotent steps with blind retries, no compensation for irreversible actions, approvals bolted on via Slack DMs with no state, and treating a workflow engine like a fancy cron with no failure handling.