57 open-source web apps: how many screens, how many end-to-end spec files

A zero means the rule found no end-to-end spec file in the repository, not that the project is untested. Many projects test in other ways, and a count of spec files says nothing about a project's quality.

If a row is wrong, write to hello@featkpr.com with the app and the column; we correct it and date the fix on this page. How corrections work.

"no count": the screens cannot be counted from the files, or no API spec is committed; hover for the reason, and the downloads carry it. "by hand": report T2, 17 Sep 2026. "script": tools/library_count.py, 28 Sep 2026.

A screen

a page a person can land on: one page file or one route declaration, never a component. In file-routed front ends (Next.js, SvelteKit, Nuxt, React Router route modules, a Vue app's views), page files minus layouts, API routes and _app/_document. Where the router is code (React Router, Vue Router, Angular, Ember, AngularJS), the route declarations in the router file the rule names. In server-rendered apps (Rails, Django, Laravel, Twig, Go templates, plain PHP), the templates that render a whole page: ones that extend a layout, or a Rails view that is not a partial, a layout or a mail.

No screen count

an app whose pages cannot be counted either way from its files (routes spread across the code, or screens generated from records) gets none, with the reason, rather than a guess.

An API operation

a (path, method) pair in the OpenAPI or Swagger spec committed to the repository, following a spec split into files through its $refs. No committed spec (GraphQL, RPC, a spec generated only at runtime or kept in another repository): no count.

An end-to-end spec file

a file of the suite the project runs against the running app from outside, through a browser (Playwright, Cypress, Puppeteer, Selenium, Protractor, Capybara system specs, Odoo tours) or over HTTP, in the directory the project names for it. Unit, component and in-process HTTP tests are not counted.

Two counts: report T2 counted 14 apps by hand on 17 Sep 2026 (BookStack at its v26.05.5 tag, the others at their default branch that day, with no commit recorded). On 28 Sep 2026 the script counted 43 more at each repository's default branch, and recounted 4 of T2's apps to check the rules against T2's figures.

Where both counts exist #

T2's hand count (17 Sep) beside the script's (28 Sep). The table and the picture use T2's; the downloads carry both.

InvenTree's API count differs because its spec is kept in another repository (inventree/schema): T2 read it there, and the script reads only the repository it counts.

• Not a sample of all open-source apps. The apps were picked as candidates for featkpr to map, plus well-known self-hosted apps across stacks. Nothing here says what share of all projects have end-to-end suites.
• Not a measure of quality. Fewer spec files than screens does not mean a screen is untested: unit and integration tests are not counted, and one spec file can cover many screens.
• Rules differ by stack. A screen in a Rails app (a template) and in a React app (a route declaration) are counted by different rules, so compare apps within a stack with more care than across stacks.
• One moment in time. Each count is at one commit. Repositories move every day; run the rule again and the number may differ.
• Rules, not a reading of every file. Each number is what its rule matched, and the 4 recounted apps are checked against T2's hand counts. A rule can still count a file it should not, or miss one.

1. Clone the repository and check out the commit in the row.
2. List the files: git ls-tree -r --name-only HEAD.
3. Apply the row's rule: the files whose path matches the pattern, minus the exclusion, or the files whose text matches it. The count is the number of files (or of route declarations, where the rule says occurrences).

The script that did it, tools/library_count.py (Python, no dependencies beyond an optional YAML reader), shallow-clones each repository with no blobs, fetches only the files a rule must read and prints one JSON record per app. Its output of 28 Sep 2026 is the receipt the site reads. Run today, it counts today's default branch, so the numbers move with the repositories.

The data are also on GitHub: github.com/featkpr/open-source-apps-counted (CC BY 4.0). The counting script follows there.

receipt: research/reports/library-count-2026-09-28.json · 47 records, one per repository counted on 28 Sep 2026

E2EGit: A Dataset of End-to-End Web Tests in Open Source Projects (Sergio Di Meglio, Luigi Libero Lucio Starace, Valeria Pontillo, Ruben Opdebeeck, Coen De Roover, Sergio Di Martino) publishes a dataset of GitHub repositories written in Java, JavaScript, TypeScript or Python that depend on a web framework. It finds Selenium, Playwright, Puppeteer and Cypress tests (and JMeter and Locust performance tests) and counts test cases per file.

Paper: doi 10.1109/MSR66628.2025.00121 (the authors' PDF). Data: Zenodo, doi 10.5281/zenodo.14221860, CC BY 4.0. Every E2EGit figure above is from the paper, read 28 Sep 2026.

• open-source-apps-counted-2026-09-28.csvUTF-8, one row per app. The first lines start with # and name the licence, the source and the dates: skip them as comments (pandas: comment="#").
• open-source-apps-counted-2026-09-28.jsonThe rows, a dictionary of every column, the summary figures on this page and the provenance.

Licence: Creative Commons Attribution 4.0 International (CC BY 4.0). Use it for anything, including commercially; name featkpr as the source and link to this page.

Cite this #

@misc{featkpr_counted_20260928,
  author       = {{featkpr}},
  title        = {57 open-source web apps counted: screens, API operations and end-to-end spec files},
  year         = {2026},
  month        = {sep},
  howpublished = {\url{https://featkpr.com/library/counted}},
  note         = {Dataset, CC BY 4.0. Counted 17 Sep 2026 and 28 Sep 2026}
}