Gallery data

The GitStats Gallery makes a gitstats report every Monday for each repository of gallery/repos.yml, and publishes what it counts as data next to its pages. These files are free to read from scripts, notebooks and other sites.

Ecosystem rankings

Each group of gallery/repos.yml (an ecosystem: Python, CNCF graduated, Apache, …) has a page that ranks its repositories by the share of the last complete quarter’s commits that name an AI coding assistant, and the same ranking as JSON at the same address:

/gitstats/gallery/ecosystems/<id>.html and <id>.json, for example python.html and python.json;
/gitstats/gallery/ecosystems/index.html and index.json: every ecosystem side by side.

The ids are those of gallery/repos.yml; an ecosystem’s page appears once one of its repositories has a report. /gitstats/gallery/sitemap.xml lists the gallery’s pages for search engines.

A commit counts as AI-assisted when it names an assistant as its author, as a co-author, or in the line the assistant signs with. The counts are lower bounds: an assistant that leaves no mark in a commit is not counted, and a mark does not say how much of the commit the assistant wrote. A commit counts in the quarter it was written (its author date).

`<id>.json`

A part of python.json, from a gallery of 24 repositories:

{
  "schema_version": 1,
  "generated_by": "gitstats 2.12.0",
  "generated_at": "2026-10-26T00:41:07+00:00",
  "ecosystem": {"id": "python", "name": "Python", "title": "Python open source projects"},
  "quarter": {"id": "2026-Q3", "label": "Jul–Sep 2026"},
  "previous_quarter": {"id": "2026-Q2", "label": "Apr–Jun 2026"},
  "summary": {
    "repositories": 6,
    "with_commits": 6,
    "naming": 4,
    "median_share": 0.0115,
    "highest": {"id": "pandas", "name": "pandas", "share": 0.6674}
  },
  "repositories": [
    {
      "rank": 1,
      "id": "pandas",
      "name": "pandas",
      "repository": "pandas-dev/pandas",
      "report": "../pandas/",
      "quarter": {"commits": 947, "ai_assisted": 632, "share": 0.6674},
      "previous_quarter": {"commits": 630, "ai_assisted": 356, "share": 0.5651},
      "commits_last_12mo": 2885,
      "active_authors_12mo": 398,
      "health": "steady",
      "last_commit": "2026-10-18",
      "assistants": ["Claude", "Devin", "GitHub Copilot"]
    }
  ]
}

index.json has schema_version, generated_by, generated_at, quarter, and ecosystems: for each, its id, name, title, page, data (the file names of its page and JSON) and its summary, as above.

Stable fields

While schema_version is 1, these fields keep their names, types and meaning; a change to any of them comes with a new schema_version:

schema_version;
ecosystem.id; quarter.id and previous_quarter.id (YYYY-Qn);
summary.repositories, summary.with_commits, summary.naming, summary.median_share, summary.highest (id, name, share, or null when no repository names an assistant);
for each of repositories, sorted by rank: rank, id, repository (owner/name on GitHub), quarter and previous_quarter (commits, ai_assisted, share), commits_last_12mo and active_authors_12mo;
in index.json: ecosystems[].id, ecosystems[].data and ecosystems[].summary.

A share is a fraction from 0 to 1, rounded to four decimals, or null when there are no commits to make one of. quarter (in a repository) is null when its report does not reach back to that quarter. The ranking: the largest quarter.share first, then the most commits in the quarter, then the name; repositories without a share last.

Everything else may change without notice: the display names and titles, quarter.label, generated_by, generated_at, report (where the repository’s report is, relative to the JSON file), health (the repository’s pace, as its report’s History page labels it), last_commit, and assistants (the assistants its commits name most). New fields may be added at any time.

AI-assisted commits

ai.json and ai.csv hold the counts of the AI-assisted commits page, month by month and quarter by quarter, for every repository of the gallery; ai/<YYYY>-Q<n>.json keeps each quarter as it was first published.

Organizations

organizations.json and organizations.csv hold the counts of the Organizations page: commits by the domain of the author’s email address, added up across the gallery’s repositories. Domains and counts only, never an address.

organizations lists every organization, the most commits in the last 12 months first: organization (its domain), commits_last_12mo (the 12 months that end with each repository’s last commit), commits (all of them) and repositories (the names of those it committed to). repositories lists, for each repository, its commits of the last four complete quarters (named in quarters) and by_kind, how they divide between organization, one_person (a domain only one person commits from, most often that person’s own), personal (mail providers anyone can sign up for), hidden (code hosts’ noreply addresses), unknown and bot; and organizations_last_12mo, its five organizations with the most commits in the last 12 months. A domain says where an address is, not who paid for the work. New fields may be added at any time.

Each of the 200 organizations with the most commits in the last 12 months has a page, orgs/<domain>.html, with its commits to each repository, and a badge that is made again with every build: orgs/<domain>.svg, and orgs/<domain>.json for a shields.io endpoint badge. In Markdown:

[![GitStats Gallery](https://shenxianpeng.dev/gitstats/gallery/orgs/google.com.svg)](https://shenxianpeng.dev/gitstats/gallery/orgs/google.com.html)

In organizations.json, each organization’s by_repository lists the repositories it committed to, each with its commits, commits_last_12mo, addresses, first_commit and last_commit.