Or: These are really good points for a visibly-user-facing post-alpha service, but isn't it a bit overengineered for an experimental internal service whose clients can tolerate the risk of occasional downtime?