Lesson learned the hard way: add observability to your pipelines before you need it, not after something breaks in production at 11pm. At minimum: log input/output record counts at each stage, track processing time per batch, and alert on anomalies. Future you will sleep so much better.
