Open data

The Open Benchmark

Live software-delivery percentiles from 200 curated, actively developed open-source repositories: cycle time, time to first review, PR size, and throughput, recomputed from the GitHub API on a rolling daily refresh. No form, no gated PDF, no stale annual report.

The numbers

Download JSON

Loading the live percentiles…

Percentiles are computed across repositories (each repo contributes its own median), so a few very large projects cannot skew the distribution.

How to read these numbers

What is a good cycle time?

Cycle time here is the median hours from a pull request opening to merging, per repository. Long cycle times usually mean work is waiting on review pickup, CI, or batching, not that anyone is writing code slowly. Full explainer →

How fast should the first code review arrive?

Time to first review (pickup time) is the median hours from a pull request opening to its first review. It is typically the largest component of cycle time, and the cheapest one to fix. Full explainer →

How big should a pull request be?

PR size is the median changed lines per merged pull request. Smaller changes review faster and revert cleaner; sustained very large PRs usually mean batched work. Full explainer →

How many PRs does an active project merge per week?

Throughput is merged pull requests per week over the sample window. It is a volume signal to read alongside the speed metrics, never a target on its own.

Reference bands from vendor reports (like LinearB's published benchmarks) describe private-company teams; this dataset shows real, current open-source distributions. The two are complements, so cite whichever matches the population you care about.

Methodology

  • Population: a curated, static list of 200 popular, actively developed open-source repositories, chosen for real ongoing development rather than stars alone.
  • Per-repo stats: each repository contributes the median over its 50 most recently merged pull requests, fetched from the GitHub API.
  • Aggregation: the percentiles above are computed across repository medians, so no single project dominates.
  • Freshness: a rotating chunk of the list recomputes every day; the full set turns over roughly every five days. The lastUpdated stamp in the JSON is the newest computation.
  • Honest caveat: open-source workflows differ from private-company workflows (volunteer reviewers, time zones, drive-by contributions). Directional context, not a verdict on any team.

Cite this data

Free to cite with attribution (CC BY 4.0). Suggested form: “DevPerform Open Benchmark, live percentiles from 200 open-source repositories, devperform.io/benchmark/industry”. The JSON endpoint is stable and needs no key.

Where does your team land?

Benchmark any public GitHub repository against these distributions in seconds. Free, no signup.

Benchmark a repo →

Or explore the full dashboard with sample data. No signup.

For private repos and continuous tracking, with the developer surveys that explain the numbers, DevPerform connects to GitHub, GitLab, and Bitbucket in minutes.