percentile

How it works

Evaluates a specific point in the data distribution, such as the 99th or 50th percentile. This provides flexibility for monitoring high-precision tail latency or median performance when standard defaults are insufficient.

When to use

• Monitoring high-precision tail latency (e.g., p99) for critical services.
• Tracking median performance (p50) to focus on typical user experience.
• Enforcing specific performance contracts that require non-standard percentiles.

When not to use

• Very small sample sizes where high percentiles are statistically unstable.
• When standard p95 or average monitoring provides sufficient coverage.

State explanations