A measurable quantity that describes one specific aspect of a service's behavior, the input to an SLO.
A Service Level Indicator (SLI) is a quantitative measure of one aspect of how a service is performing, request success ratio, p95 latency, queue depth, freshness lag. An SLI is just a number: 'over the last hour, 99.94% of /checkout requests returned 2xx within 500ms.' SLIs are the inputs to SLOs (which set a target for an SLI) and to error budgets (which measure how much of that target you've burned).
Most teams pick the wrong SLI to measure. Common mistakes: using percent CPU as a proxy for user pain (it isn't), or measuring at the wrong percentile (averages hide tail latency). A good SLI maps tightly to user-perceived experience; a bad one is just a metric that happened to be easy to graph.
See the part of the platform that handles sli (service level indicator) in production.