Percentile Metrics (Median, P75, P90) Explained

Typo surfaces key engineering metrics using median (P50) and percentiles (P75, P90) instead of relying only on averages. Percentiles describe how values are distributed across real work, which makes them more reliable for understanding delivery behavior, variability, and risk.

Most software delivery metrics are skewed by nature. A small number of very slow items can distort averages and hide meaningful signals. Percentiles avoid this problem by showing how most work behaves and where delays accumulate.

What Do These Metrics Mean?

Median (P50)

The midpoint of the dataset. 50% of values are faster, and 50% are slower.

In practice, this represents the typical experience of work flowing through the system. It is stable and not influenced by extreme cases.

75th Percentile (P75)

75% of values fall at or below this point. 25% take longer.

This captures the upper bound of normal behavior. It shows how long most work takes once mild friction is included.

90th Percentile (P90)

90% of values fall at or below this point. 10% take longer.

This exposes the long tail. These are the cases that typically lead to frustration, missed expectations, and planning risks.

Where Percentiles Apply in Typo

Cycle Time Coding Time Pickup Time Review Time PR Size Avg. Commit/PR

How to interpret percentiles in cycle time

Consider all PR cycle times for a team, sorted from fastest to slowest.

• P50 (Median) Half of PRs finish faster than this. This is the team’s typical delivery speed. If P50 moves, the day-to-day flow has changed.

• P75 Three out of four PRs finish within this time. This illustrates where most work is concentrated once normal friction is taken into account.

• P90 Only 10% of PRs take longer than this. This is the long tail—blocked, complex, or poorly scoped work.

How to read patterns

• P50 stable, P90 rising → core flow is fine; edge cases are breaking.

• P50 and P75 rising → broad slowdown across the team.

• Large gap between P50 and P90 → high variability and inconsistent flow.

Why Typo Emphasizes Percentiles

Percentiles help engineering leaders:

• See real distribution instead of averages

• Detect early risk before outcomes degrade

• Separate widespread issues from edge-case failures

• Prioritize fixes that improve the experience for most developers

They make delivery performance observable, interpretable, and actionable.