A payment notification service on our EKS cluster failed silently for 47 minutes — PagerDuty never fired, Loki never alerted, we found out from a customer tweet. The root cause was a microservice extraction that lost the structured logging config. The fix wasn't more tooling. It was treating observability as a hard deployment requirement, enforced with an Open Policy Agent policy in CI. Here's the Rego, the GitHub Actions gate, and the before/after of what changed.