---
title: "P50, P95, P99 & Variance"
description: "Turn per-request timings into a latency report: the P50, P95 and P99 grid, the shape behind each row, tail exposure, and how many runs a claim needs."
canonical_url: "https://fanout.sh/inference-eng/curriculum/04-05"
md_url: "https://fanout.sh/inference-eng/curriculum/04-05.md"
access: "pro-preview"
---

# P50, P95, P99 & Variance

Turn per-request timings into a latency report: the P50, P95 and P99 grid, the shape behind each row, tail exposure, and how many runs a claim needs.

- Access: Fanout Pro (public preview)

## Public preview

The record from last lesson ends with a rule about order: keep the raw per-request data, because summaries come last. This lesson is the summarizing. The 199 kept requests of the course capture become the handful of numbers a report can print, and the machinery for that comes from sampling and statistics, which built percentiles on twenty synthetic timings and…

## In this lesson

- The report grid
- The shape behind each row
- Tail exposure
- Population choices
- Sample size and run planning

## Continue learning

Pro unlocks the complete lesson, video, practice, and code where available.

---
This representation contains public Fanout content only. Protected Pro lessons, account data, billing, checkout, and pricing are not included.

Browse the public content map: https://fanout.sh/sitemap.md