A calm, searchable map of the questions humanity has not answered yet.
The Open Questions Atlas turns open problems into a place to wander. Search by phrase, choose a region of the atlas, follow a curated trail, or choose Surprise me and let a non-repeating question find you.
Live site: purysho.github.io/OpenQuestionAtlas
- Atlas regions — browse the unknown by field, from cosmology to medicine.
- Curated trails — follow a short thread through origins, time, or the mind.
- Question of the week — begin with one carefully chosen open problem.
- Related questions — keep following a line of inquiry without leaving the atlas.
- Shareable links — every question and trail has a stable URL you can send to someone else.
- Add to sky — every question card has an Add to sky ✦ button that opens
Lodestar with the question pre-loaded as
a star (tagged
origin: "atlas"), so a question that stayed with you lands in your personal night sky.
The Atlas is a stateless discovery tool; Lodestar
is the memory layer. Add to sky encodes the question into a small payload in
the Lodestar URL hash (…/lodestar/#/add?s=<encoded>) — no backend, no shared
storage, the URL is the whole wire. The shared payload + encoding contract lives
in src/lib/starLink.js, copied verbatim from Lodestar
and covered by a round-trip test so the sender and receiver never drift.
Want to help grow the map? Open an issue with the Add a sourced open question template, or read CONTRIBUTING.md.
npm install
npm run dev
npm testNode 20 is the supported runtime. The development server prints the local URL.
The browser fetches only static JSON under public/data/. index.json is the
manifest: the loader fetches every dataset it lists, validates entries against
schema.json, drops and logs invalid or unsourced entries, merges the valid
records, and removes duplicate IDs. No backend, database, API key, or external
runtime API is involved.
That manifest is the seam between today's curated seed and future source imports. Adding another static dataset should require a new JSON file plus one manifest entry; the React application should not change.
Most knowledge products organize answers. This one maps the edges of knowledge instead. Each question is treated as the hero, with enough context to understand why it remains open and a direct path into the source material.
Every shipped question must be traceable. A smaller honest atlas is better than a larger one padded with invented questions, authors, or URLs.
The manifest seam also lets a build-time pipeline add whole datasets from
permitted external sources. scripts/ingest/ fetches a source, reformats it
into the schema, and writes static JSON to public/data/ plus an updated
index.json and LICENSES.md. The app is untouched — it just sees more files.
Adding a source is: confirm its license, write one adapter, register it, and run
npm run ingest. Full instructions, the adapter contract, and the rules
(license gate, no fabrication, and the Wikenigma link-don't-ingest / no-scrape
policy) are in scripts/ingest/README.md. The
Erdős Problems dataset is the worked example.
npm run devstarts Vite.npm testruns the loader and discovery tests once.npm run lintruns the repository's source/build validation gate.npm run buildcreates the static production site indist/.npm run ingestruns the build-time ingest pipeline (see above); it is never part of the app or the deploy.
Question curation is the heart of this project. Read CONTRIBUTING.md before adding or changing data. In particular, use an approved primary source, record its real URL and license on every entry, and run all three gates.
Good first contributions include adding a small sourced batch in an underexplored field, proposing a curated trail, or improving a question's tags. The repository issue form walks through the source and licensing information needed for each proposed question.
Application code is available under the MIT License. Question data keeps its source-specific terms; see LICENSES.md.