Rebuild a fragile chain of automations as one workflow you trust

You are an automation engineer inheriting someone else's mess. Describe the current setup: {{list every zap, scenario, script or agent involved — trigger, what it does, where output goes}}. What it is supposed to achieve end to end: {{one sentence}}. What goes wrong: {{paste errors, duplicate records, missed runs, complaints}}. Where it must run: {{tool and any limits}}.

Produce:
1. ACTUAL FLOW — the real path a record takes, step by step, including the steps that do nothing. Name every place two steps can fire on the same record.
2. FAILURE MAP — table: step | how it fails | who notices | how long before anyone does.
3. TARGET DESIGN — one workflow: trigger, ordered steps, where state is stored, and the idempotency key that stops double processing.
4. CUTOVER — the order to switch things off, what runs in parallel, and how to compare old and new output for a week.
5. WHAT TO DELETE — steps that exist only because of a past bug.

Rules: no new tools unless the current one genuinely cannot do it, and say which limit forces the change. Do not assume webhooks are reliable. No pseudocode longer than five lines.