unsolved.now

A public catalog of the great unsolved problems of science and engineering — how long each has been open, why it resists, what progress has been made, and what falls when it's solved. Every date, prize, and progress claim is cited to an authoritative source.

Live: https://unsolved.now

Stack: zero-plugin Jekyll on the native GitHub Pages branch build

Decision criteria, in order: boring/durable, data-driven, minimal build complexity, fast. The choice:

• Jekyll, built natively by GitHub Pages from main. Pushing to main is the entire deploy pipeline — no Actions workflow, no Node toolchain, no lockfile that can rot. GitHub has run this exact path for a decade and maintains it; it will still build cleanly in two years untouched.
• Zero Jekyll plugins. Meta/OG tags, sitemap.xml, feed.xml (Atom), and robots.txt are small hand-written Liquid templates — nothing to break, exact control over the SEO surface.
• Data-driven: every problem is one Markdown file with YAML frontmatter in _problems/; all pages render from that data. No hardcoded catalog HTML.
• Fast: CSS is inlined (~9 KB); one self-hosted, subsetted chalk font (Caveat, 18 KB woff2, latin-only, weight-pinned) used sparingly for display; JS is ~20 lines of optional progressive enhancement.

Design

Chalkboard: the universal image of an unsolved problem is a blackboard covered in half-finished equations. Dark board (#1B2521), chalk-white serif body, a handwritten display face (Caveat) reserved for the emotional beats — the big "open for N years" counters and the timeline years — and a single warm chalk-yellow accent. Fully functional without JavaScript; keyboard-navigable; respects reduced motion.

Data schema

One file per problem: _problems/<slug>.md. Frontmatter fields:

Field	Required	Notes
slug	yes	Must match the filename (drives the URL /problems/<slug>/).
name	yes	Short proper name ("The Riemann Hypothesis"). Shown as the eyebrow and in rows.
title	yes	Curiosity-first, plain-language hook — what we don't know.
field	yes	A key from fields in _config.yml.
statement	yes	One precise sentence of the actual open question.
first_posed	yes	Year (integer). Drives the "open for N years" counter.
posed_by	no	Who posed it / where it originates.
prize	no	Real, verified prize (e.g. "Clay Millennium Prize — $1,000,000"). Omit if none.
summary	yes	3–5 accessible sentences. Doubles as the meta description.
why_it_matters	yes	2–3 sentences of concrete stakes.
progress	no	List of "YEAR — milestone" strings, real and dated.
references	yes	List, min 2: url, source, note. Independent, authoritative, verified live.
status	yes	curated.
date_curated	yes	YYYY-MM-DD; also the "verified still open as of" date.

The "open for N years" counters are computed at build time from first_posed and the build date — they stay current on every rebuild with no manual edits.

Content curation rules

1. Zero fabricated data. Every date, name, prize, and progress claim traces to a linked authoritative source, verified live (HTTP 200 + real content) at curation time.
2. Must be genuinely open. Each entry is checked for recent resolutions before inclusion (problems do fall — the 3D Kakeya conjecture fell in 2025). The date_curated doubles as "verified still unsolved as of".
3. Curiosity-first, precision underneath. Titles are plain-language hooks; the statement and details are exact. No jargon in the hook, no hand-waving in the body.
4. Authoritative sources only — Clay Institute, arXiv, Nature/Science, CERN, NASA, Nobel, university/institute pages, Quanta for recent progress. Wikipedia allowed as at most one of the two links.
5. If it can't be sourced or might already be solved, it's cut.

Local development

brew install ruby@3.3
export PATH="/opt/homebrew/opt/ruby@3.3/bin:$PATH"
bundle install
bundle exec jekyll serve   # → http://127.0.0.1:4000

One-command build check: bundle exec jekyll build.

Deployment

Push to main. GitHub Pages builds natively (the pages build and deployment workflow appears in the Actions tab). Custom domain unsolved.now via CNAME. Verify live after deploy:

curl -sL -o /dev/null -w "%{http_code}" https://unsolved.now/
curl -s https://unsolved.now/sitemap.xml | head
curl -s https://unsolved.now/feed.xml | head

Owner TODO

• Follow-the-board endpoint — create a Buttondown/Formspree form and put its POST URL into form_endpoint in _config.yml. Until then the follow CTA is a mailto: to contact_email (currently the owner's Gmail — change it there if you'd rather not expose it).
• Analytics (optional) — GoatCounter is a privacy-friendly, no-cookie-banner option. Deliberately not integrated with a placeholder key.

Everything else (GitHub Issues for suggestions, gh auth, www DNS, Ruby TLS) is resolved and verified.

History

The first iteration cataloged unsolved workflow problems (a demand-discovery index). That version is preserved on the workflow-catalog branch. The main site pivoted to the great open problems of science and engineering.