Service Level Indicator
also called SLI
The actual measurement of a service's behaviour that an objective is set against — a ratio of good events to valid events.
The definition worth internalising is the ratio form: good events divided by valid events. "Requests that returned a non-5xx within 300 ms, divided by all requests that reached the service." That shape makes an SLI directly comparable to a target and directly convertible into an error budget.
Three decisions determine whether an SLI is meaningful. What counts as good — include latency, not just success, or a service that is up and unusably slow scores 100%. What counts as valid — excluding client errors is usually right, excluding a whole category of traffic because it looks bad is not. Where it is measured — at the load balancer, at the service, or in the client. The closer to the user, the more honest, and the more it includes things you do not control.
Most services need two or three: availability, latency, and often correctness or freshness. A dozen SLIs means none of them is driving a decision.