A free, open-source, deeply interactive learning platform. Learn by doing — without the ceiling.
Brilliant proved that short interactive problems beat reading. It also charges a lot and stops well short of the material a practitioner actually needs. This keeps the interaction model and removes the ceiling: every topic runs from foundation to frontier on one page, and you choose the depth.
Live: https://entangledquantum.github.io/anthropocene/
- Predict → run → confront. You commit to an answer before the simulation runs. Watching your prediction fail is the mechanism; a paragraph cannot do it.
- Nothing asks you to write code. Nobody hand-writes a solver in 2026, so testing that tests the part of the skill that is now free. Instead you sketch the curve you expect and watch the real one land on your guess, rank methods by cost, or sort a symptom into cause — the judgement a model will not exercise for you.
- Visualizations that could not be a static figure. Stability regions evaluated per pixel on the GPU that you drag a point around; phase-space area transport that makes Liouville's theorem something you watch rather than accept.
- One concept, one owner. A concept is taught by exactly one lesson, enforced at build time. Everything else links to it — so the material never repeats itself, and the list of unwritten topics is generated rather than maintained.
- Foundation to frontier in one lesson.
<Tier>blocks fold away the depth until you ask for it. - Recall built in. Cards are authored inline where the idea appears and scheduled with FSRS-6.
- Yours, locally. XP, streaks and review state live in SQLite in your browser. No account, no server, no subscription, works offline.
| Framework | Astro 7 + React islands — 0 JS on prose, hydration per widget |
| Math | KaTeX rendered at build time via Satteri's native math support |
| Plots | Canvas data layer + SVG axes on d3-scale / d3-shape |
| GPU visuals | Raw WebGL2 fragment shaders (src/components/viz/gl/) — stability regions, phase fields, the landing-page Lorenz attractor |
| Storage | SQLite-WASM (opfs-sahpool VFS — no COOP/COEP, so it works on GitHub Pages) |
| Recall | ts-fsrs (FSRS-6) |
| Search | Pagefind — chunked index, downloads only what a query needs |
| Python | Pyodide, lazy, opt-in per lesson |
npm install
npm run dev # http://localhost:4321/anthropocenenpm run content:check # concept-graph rules, gaps, per-lesson requirements
npm run content:gaps # the generated to-write queue
npm test # asserts the physics: orders, stability, energy drift, area preservation
npm run build # runs content:check first
npm run screenshot # regenerate docs/landing.png after frontend changesRead AGENTS.md. It is the formula — the no-code-writing rule, the bar a visualization has to clear, the widget vocabulary, and the architectural constraints that will otherwise cost you an hour each.
LEARNING-PLAN.md is the queue: what to write next, in order, basic → advanced. Ship a lesson, delete its entry in the same commit.
npm run new:lesson -- --path computational-physics --chapter 03-ode-solvers \
--id 05-dormand-prince --title "Adaptive steps in practice" \
--teaches embedded-pairs --requires rk4,butcher-tableauIf the path does not exist, it is created. Give it a category and it builds the learning path around it.
content/ is fully separate from src/. Nothing in it imports app code and nothing in
src/ hardcodes a lesson.
The first path is Computational Physics: floating point → finite differences → ODE solvers one at a time → structure-preserving integrators → spectral, variational, Monte Carlo and differentiable simulation.
MIT for the code. Content under CC BY-SA 4.0.