ScholarLib

A small, opinionated library of CS / economics papers I've actually read or want to read, with a VitePress site on top so I can find them again.

Live site → https://badhope.github.io/ScholarLib/

What's in it

• 503 papers currently indexed across 33 Computer Science topics and 18 Economics & Management fields
• All 503 DOIs are real (verified against doi.org); the entries are sourced from arXiv, NeurIPS, ICML, JMLR, Nature, AER, JFE, JEP, and other top venues
• 1 open-access PDF bundled (sample reference); the rest link to the publisher via DOI
• Each paper has a metadata.json and a .bib you can copy straight into LaTeX
• 8 publication year buckets (2019-and-earlier, 2020 – 2026) wired into the folder structure
• A static site that builds into a searchable, dark-mode-friendly doc

I originally kept this as two separate repos (CS and Econ/Management) and a pile of Google Docs bookmarks. This repo is the merge. Contributions for additional CS or EM papers are very welcome.

Run it locally

npm install
npm run docs:dev      # http://localhost:5173

The Python helpers in scripts/ only matter if you're adding or auditing papers:

python3 scripts/validate_papers.py     # CI runs this
python3 scripts/recover_pdfs.py        # one-off, fills in missing PDFs from arXiv

A few choices worth flagging

• VitePress, not Next.js / Astro. Static output, no JS framework on the page, and VitePress' default theme is already 90% what I'd build by hand. The build is ~5s.
• metadata.json per paper, not a single index. Lets each paper's folder be self-contained — easy to vendor a topic out, easy to PR a single paper. The downside is the auto-generated sidebar has to be rebuilt after structural changes; build_sidebar.py does that and writes into docs/.vitepress/config.ts.
• The site is multilingual (en / zh), but the paper content is English-only. I only localized UI strings. Translating the abstracts is a different project and not on the roadmap.
• PWA is opt-in. The service worker only registers in production builds and on load, so dev HMR isn't disturbed.

Things I deliberately didn't do

• No Algolia. The built-in local search is fast enough for a few thousand docs and I don't want to pay for it.
• No comment system. The repo is read-only for most people; the GitHub issue template is enough.
• No analytics.
• No telemetry on the PDF recovery script. If it pings arXiv 50 times you might get rate-limited; that's the worst case. (The current indexed papers are journal-only, so the script is mostly a no-op — see the DOI line in metadata.json.)
• No bundled PDFs. Each entry links to the DOI; the repo stays small and respects publisher paywalls. Run scripts/recover_pdfs.py only if you actually want to mirror a paper locally.

Repo layout

.
├── docs/                       VitePress site
│   ├── .vitepress/             config, theme overrides
│   ├── {index,about,architecture,citation-graph}.md
│   ├── zh/                     Chinese UI strings
│   └── public/                 static assets, PWA manifest, sw.js
├── papers/                     the actual content
│   ├── computer-science/<topic>/
│   └── economics-management/<year>/
├── scripts/                    build-time helpers (Python)
├── tests/                      unit tests for the scripts
├── templates/paper-template/   copy this when adding a new paper
├── .github/workflows/          CI, deploy, security scans
├── build_sidebar.py            auto-generates the VitePress sidebar
├── postbuild_a11y.py           runs after vitepress build
├── Makefile                    the day-to-day commands
└── package.json

Contributing

Bug reports and paper additions are the most useful. See CONTRIBUTING.md.

If you're proposing a new paper, please use the new_paper issue template — it asks for the metadata fields the validation script needs, and submissions missing them just bounce.

License

MIT for the repo and the metadata. The papers themselves keep their own licenses — check each entry's metadata.json before redistributing.

---

Maintained by @badhope. The site rebuilds on every push to main.