The time between a request starting and a response arriving, the user-perceived 'is it fast' metric.
Latency is the elapsed time between a request being sent and a response being received. It's measured at percentiles (p50, p95, p99) rather than averages because averages hide the tail: a service can have a 50ms average and a 2-second p99, and the p99 is what 1% of users feel. SRE teams set SLOs on latency at high percentiles because tail latency, not average latency, drives churn.
Latency is the most directly user-perceived signal in the golden-signals set. A slow service feels broken even when error rates are zero. Latency SLOs that include the percentile (e.g. '99% of /checkout responses under 500ms') are dramatically more useful than average-latency targets, because they constrain the tail that customers actually experience.
See the part of the platform that handles latency in production.