Reliability does not begin with an uptime dashboard. It begins with the design of the system: how requests fail, how work is retried, and how an operator knows what happened.
For each critical workflow, ask what can go wrong and what the user will see. Put limits on work, make retries deliberate, and avoid hiding failure behind an interface that appears successful.
Use monitoring to answer real operational questions. A useful signal tells you which workflow is affected, what changed, and where to look next. Recovery should be a documented part of the system rather than an improvised response.