Decisions

Short records of choices that shape the runbooks — what I chose, what I rejected, and what it costs.

Decision	Choice	Why
Two tracks	Self-hosted k3s and managed AWS/EKS, deliberately with different tools	Running Argo CD and Flux, Kyverno and Gatekeeper shows what each actually costs
One application	The same small Spring Boot service on both tracks	A second app teaches nothing; the comparison is the point
Order	AWS first, then Terraform, then Kubernetes on metal, then EKS	Rebuild AWS as code while it’s fresh; learn Kubernetes where nothing is hidden
Secrets on the homelab	OpenBao instead of HashiCorp Vault	Same API, open-source licence (MPL-2.0)
Evidence	Every runbook ends with checks that must pass and a written record	“It applied” isn’t evidence; a timed rebuild is