About

A Data Science knowledge base with three rules, each of them enforced by the build rather than by good intentions. This page says what the rules are, who wrote the site, and what you may do with it.

What the build refuses to publish

• A page whose claim is not attributed to a source someone actually opened — at the section cited, not the paper in general.
• An English concept page with no figure. `NO-FIGURE` is an error there, not a warning.
• A worked example whose output was not produced by running it. Every number printed on this site came out of a script in `scripts/`.
• A page sourced entirely from library documentation. Something has to own the idea: a textbook or the primary paper.
• A colour pair below the contrast bar, a still frame that selects nothing, or a figure that ships more JavaScript than the budget allows.

Every page also lists its own sources at the bottom. See the sourcing rules in full.

What it is not

It is not comprehensive, and it is not a course. Pages exist because a specific idea was worth getting right, not because a syllabus called for them. Where a result did not come out the way it was expected to, the page says so first — several pages exist mainly to record that the tidy version of a story is wrong.

Who wrote it

Marginhand. The name is the manicula — the small pointing hand a medieval reader drew in the margin to mark the exact passage worth coming back to. That is the habit this site is built on: not citing a work, but citing the place in it, and checking the sentence is really there.

Licence

What anyone can read here is free to reuse. The free pages in full, the open part of every other page — the definition, the formula, the intuition, the sources — and the catalogues are CC BY-SA 4.0: use them, change them, publish them, including commercially, as long as you credit Art of DS with a link and pass the same freedom on. The parts that come with lifetime access are all rights reserved: read them, learn from them, quote a few lines with a link, but please do not republish them. The code in the worked examples is MIT wherever it appears: take it and use it. Quotations from other people’s papers and books are theirs, not mine, and no licence here grants you anything in respect of them — which is exactly why every one of them is cited.

Colophon

Next.js and MDX, with a content layer that validates every page against a schema before the build will touch it. Figures are hand-built inline SVG — no chart library — so each one can say the specific thing it is there to say. Search is a small JSON index for titles and terms, and Pagefind over the built HTML for everything else. Type is IBM Plex. Visits are counted with Umami, which sets no cookie and keeps nothing that could tell who you are; a browser that asks sites not to track it is not counted.