Why on-call does not scale, where the minutes in an incident actually go, and how to move from a rota to an always-on reliability function — with autonomy you can bound, approve and audit.