Production-grade error handling in .0n workflows

I see a lot of people writing .0n workflows that work great in testing but break in production. Here are patterns I've learned the hard way: **1. Always validate inputs before API calls** Don't assume the data shape. Use a validation step. **2. Use retry with exponential backoff** APIs fail. Networks flake. Your workflow should handle it: ```yaml retry: max: 3 backoff: exponential initial_delay: 1000 ``` **3. Log everything to the audit trail** When something breaks at 3am, you'll thank yourself. **4. Use conditional steps, not separate workflows** One workflow with branches > five separate workflows. **5. Set timeouts on every external call** A hanging API call will block your entire pipeline. What patterns have you all found useful?