A platform engineering walkthrough covering Sveltos ClusterPromotion for progressive rollouts across multiple Kubernetes clusters. The author shares a real architectural mistake — a circular dependency between ArgoCD and Sveltos — where ArgoCD managed Sveltos resources and Sveltos managed ArgoCD, causing a cascading failure when a repo restructure triggered ArgoCD to prune the ArgoCD ClusterProfile. The fix: a bootstrap layer where platform tools are installed once manually, outside the GitOps loop. The post then dives into ClusterPromotion, showing how a single manifest defines a staged rollout (dev → staging) with Lua-based health checks, soak periods, and automatic promotion gates. Key details include how Sveltos creates per-stage ClusterProfiles internally, how health checks distinguish 'deployed' from 'healthy', and known gaps like no native Prometheus/Datadog metric gating, no immutable audit trail, and no Cosign image verification.
