Observability programs often begin by collecting everything and end with teams unsure which dashboard matters. Start instead from the decisions an operator must make during a failure.
Signals should lead to actions
A useful alert names the affected user journey, identifies an owner, and links to a tested response. If a signal cannot change a decision, it is probably diagnostic data—not an alert.