Cloud SLAs promise uptime statistically, not absolutely. This article shows how to treat observability as a control plane that detects regional degradation, triggers automated failover, and confirms recovery rather than paging a human. Covers active-active architecture, DNS TTL pitfalls, composite Prometheus alerting, Route 53 weighted routing, and confirmation loops backed by a real S3 outage. By Saurabh Ahuja