---
license: mit
language:
- en
pretty_name: install-graph
size_categories:
- 1K<n<10K
tags:
- npm
- software-supply-chain
- dependencies
- node-modules
configs:
- config_name: projects
  data_files:
  - split: train
    path: out/projects.json
- config_name: frontend_a
  data_files:
  - split: train
    path: out/packages-frontend-a.json
---

# install-graph

Author: Dmitriy Semenkevich (dimhold)

This dataset measures what is physically installed in `node_modules` for 5 real projects, walked on disk rather than read from a lock file.

Headline, from the repository: "In a working Angular front end, 532 of the 1,729 installed packages are second and third copies of a name already present. They take 52.2 MB and 7,904 files", 14.4% of everything in `node_modules`. Nothing in this dataset was generated by a model.

This is a mirror. The canonical home is the Zenodo record and the GitHub repository; criteria, limitations and full tables live there.

- Paper: https://dimhold.by/papers/install-graph/
- Code and full write up: https://github.com/dimhold/install-graph
- DOI: https://doi.org/10.5281/zenodo.22128826

## Files

| path | what it is |
|---|---|
| `out/projects.json` | one row per project: installed packages, distinct names, duplicate copies and bytes, depth |
| `out/results.json` | the joined analysis output |
| `out/packages-*.json` | every installed package per project with version, nesting, bytes and files |
| `out/all-names.txt` | the distinct package names across projects |

Files ending in `.jsonl` here are `.ndjson` in the GitHub repository, renamed so the viewer reads them; the content is unchanged.

## How it was produced

`scan.mjs` walks `node_modules` on disk and `analyze.mjs` joins the scan with registry metadata. The counting rules and the disproof condition were committed before the first number (`CRITERIA.md` in the repository). Not mirrored here: `out/maintainers.jsonl`, the per package npm registry pull with the public account names of about 600 maintainers; it is third party registry metadata and stays in the GitHub repository. The scripts that turn these files into every number are in the GitHub repository. One of the 5 scanned projects is a private repository; its name is written as `cli-tool` everywhere, including `out/packages-cli-tool.json`. Its counts are unchanged.

## Licence

MIT, the licence of the repository, covering code and data.

## Citation

```bibtex
@misc{semenkevich2026installgraph,
  author = {Semenkevich, Dmitriy},
  title  = {install-graph: a seventh of node_modules is the same package, twice},
  year   = {2026},
  doi    = {10.5281/zenodo.22128826},
  url    = {https://doi.org/10.5281/zenodo.22128826}
}
```