The 99th percentile of response times, the value where exactly 1% of requests are slower, the SLO that captures tail-of-distribution pain.
p99 latency is the 99th percentile of response times for a service over a window. If 99% of requests finished under 380ms during the last hour, your p99 is 380ms. The 1% slower than that, the tail, is what disproportionately drives user dissatisfaction: in a long session, even a single very-slow page tarnishes the whole experience. SLOs typically constrain p95 or p99 (sometimes p99.9 for premium services) rather than averages, because averages hide the tail.
Average latency is the single most misleading metric in observability. A service can have a 50ms average and a 4-second p99, with users complaining and dashboards looking healthy. Switching all latency reporting to percentiles, and setting SLOs at p95/p99, is the single highest-leverage observability change most teams can make in a quarter.
See the part of the platform that handles p99 latency in production.