Eighteen months on, Harborline's stack has outlived two service rewrites and the Kubernetes move — and what kept it alive was practice, not tooling. This closing chapter names what goes wrong when those practices are skipped, orders the climb for teams starting from zero, turns the stack into a deploy safety net, and then walks one Saturday-morning checkout request through the entire pipeline as proof of what it can now answer.
5 topics
Every tool in this book is installed and working, which is exactly when observability starts to fail differently: dashboards that multiply into a graveyard, alert channels everyone has muted, three logging systems where one belongs. This chapter catalogs the anti-patterns — deployments that technically work and answer nothing — and pairs each with the corrective practice that keeps a stack honest for years, not quarters.
Then it turns forward: the maturity ladder that orders adoption rung by rung, the brownfield playbook for instrumenting six inherited services from the edge inward, and the deploy safety net where canaries are judged by RED metrics and rollbacks fire on SLO burn. The book closes where it opened — 09:14 on a Saturday morning, one checkout request — followed through every signal to show what Mara's team can now answer in seconds that Chapter 1 could not answer at all.
The observability maturity ladder — each rung needs the one below it