The Built with this chapter duplicated the landing page; a single line now points to the Research Article Gallery. Template variants describes the article layout (hero, meta columns, line TOC, margin sidenotes, heading anchors, banner, debug grid and _grid.css tokens), how to switch to paper, and uses the CLI --layout flag. Unused screenshots are removed.
fix(demo): drop embeds and references whose data is not in the repo
d3-benchmark and d3-line-quad were not used by any chapter and their data files were removed long ago. The vibe-coding-charts chapter referenced d3-hf-citations.html, which was never committed.
feat(banner): isometric GPU cluster banner for the showcase article
Port of the landing site's gpu-cluster-isometric-banner (two-layer draw, SPEED 0.65, centred on its projected bounds) as banner-gpu-cluster.html on the Embed.mount contract, mounted eagerly. It reads the nearest data-theme so the /banners gallery shows both themes. Constellation, manuscript, groove and tapestry stay available.
chore(banner): regenerate article anatomy without the content experiments
Constellation, manuscript, groove and tapestry now use the embed runtime (resize: size, eager mount, ctx.animate, ctx.colors, ctx.describe, labelColor for caption text) instead of their own bootstrap, observers and D3 loader. The hero box no longer forces an aspect ratio: the banner root sets it, as the banners skill and check:embed expect. The /banners gallery loads the runtime and shows the default banner too.
chore: import local WIP (embed runtime, check:embed, skills)
Imported from the uncommitted work of the main tree: Embed.mount runtime (public/scripts/embed-runtime.js), check:embed script, embeds migrated to the runtime, write-mdx-content and design-dataviz skills, create-html-embed skill rewrite, agent docs. The old TableOfContents.astro tweak is dropped since the component was replaced.
chore: move banned-books experiments to exp/banned-books
PEN America's index has no theme or reason column, so the infographic gains
a "what and why" section built from three separately labeled sources:
- PEN America's own thematic findings per school year (2021-22, 2022-23,
2023-24 reports), each with its own base (titles or ban instances)
- reasons cited by ALA for its most challenged books of 2021 to 2023,
grouped into tags by keyword and linked to each title's PEN ban count
- Open Library subject headings for the 1,000 most banned titles (893
matched, 59% of bans), grouped into themes by keyword, cached on disk
The section leads with the contrast between "sexually explicit" claims
(30 of 33 ALA listings) and PEN's finding that 13% of 2023-24 banned
titles describe sex on the page.
Co-authored-by: Cursor <cursoragent@cursor.com>
Add three-banned-books.html, a three.js (r170, jsDelivr) tile map of US
states where each state is a pile of books, one book per 100 school book
bans from PEN America's index. Books are colored by ban status or by
school year and drop in at a constant rate, so the tallest piles keep
growing after the others stop.
- orthographic camera so equal counts have equal on-screen heights,
a ruler every 1,000 bans next to Florida and one scale for all years
- direct labels on the three tallest piles, hover, tap and keyboard
tooltips, year buttons, play, reset view and a text summary
- instanced meshes, DPR capped at 2, rendering paused offscreen or when
the tab is hidden, no motion under prefers-reduced-motion
- build script now exports ban status per state and school year
- /infographic-3d dev-only preview (redirects to / in production)
Co-authored-by: Cursor <cursoragent@cursor.com>
Add d3-banned-books.html, an infographic of US school book bans built from
PEN America's Index of School Book Bans (2021-22, 2022-23 and 2023-24
school years, 15,940 bans): headline figures with a school-year filter,
monthly timeline with data-driven annotations, tile-grid state map, most
banned titles, authors and districts, and ban status / origin shares.
- scripts/build-banned-books-data.mjs downloads the public PEN sheets,
harmonizes their labels across years and writes a 21 KB aggregate
- /infographic is a dev-only preview at article and wide width, in both
themes (redirects to / in production)
Co-authored-by: Cursor <cursoragent@cursor.com>
feat(banner): add groove and tapestry banners, drop ridges
- banner-groove: the article pressed into a record, one spiral groove where every word is a stretch as long as the word and colored by kind, chapters as tracks separated by empty grooves, a label split by chapter share and a needle reading it as it turns
- banner-tapestry: sections as warp threads colored by chapter, weft bands for prose, code, components and tables/math; a thread floats over a band in proportion to its section's share of that kind, with twill binding and swaying fringe
- remove banner-ridges
Co-authored-by: Cursor <cursoragent@cursor.com>
feat(banner): use the constellation banner on the showcase article
Add a dev-only /banners page showing every banner at hero size in light and dark themes (redirects to / in production builds).
Co-authored-by: Cursor <cursoragent@cursor.com>
feat(banner): three data-driven banners built from the article's own anatomy
Add scripts/generate-article-anatomy.mjs (run as prebuild and via npm run banner:data) which tokenizes article.mdx and its chapters into article-anatomy.json: per-word kind and length, chapters, headings, component instances, citations and internal links.
New canvas embeds reading that file:
- banner-constellation: chapters, sections, components and citations as a star graph in reading order, every word as dust around its section
- banner-ridges: one ridgeline per section showing its word-length distribution
- banner-manuscript: the whole article typeset in microscopic columns, one bar per word colored by kind
All three are DPR-aware, theme-aware, use the primary-based palette, animate a slow reading cursor, pause offscreen and render a static frame under prefers-reduced-motion.
Co-authored-by: Cursor <cursoragent@cursor.com>
docs: document layout tokens, debug grid and new frontmatter fields
Wire Hero, MetaColumns, Toc and DebugGrid into the page, move main onto the named-line grid (wide and full-width blocks, HtmlEmbed wide), single-column print through the tokens, and remove the old hero, TOC and 3-column CSS. The paper template keeps its own hero on a single-column grid.
BREAKING CHANGE: HeroArticle.astro and TableOfContents.astro are removed. tableOfContentsAutoCollapse is now an alias of tocExpanded: false.
feat: add template preview cards and --template CLI flag
- Add `links` frontmatter field for paper template hero (Paper, Code, Demo, Data buttons)
- Fix inline meta display for single affiliation (missing CSS rule on <p>)
- Update docs: README, getting-started.mdx, article-frontmatter skill
Made-with: Cursor
feat: agent-friendly project structure and paper template redesign
- Add .ai/skills/ as generic IDE-agnostic skill location (create-html-embed, article-frontmatter, deploy-to-hf-spaces)
- Add .cursor/skills symlink to .ai/skills/ for Cursor auto-detection
- Add GEMINI.md symlink to AGENTS.md for Gemini CLI support
- Update AGENTS.md with skill table, deploy section, and README tag warning
- Update CLAUDE.md to reference .ai/skills/ paths
- Update .gitignore to track .cursor/skills symlink
- Update scaffold CLI: preserve AGENTS.md, CLAUDE.md, GEMINI.md, and .ai/skills/ in generated projects
- Redesign paper template: inline meta layout, hero banner after meta, centered h2 with decorative rules
- Add gallery link and live test Space links to documentation
- Remove AUDIT-2026.md, ROADMAP-2026.md, CHANGELOG.md from tracked files
Made-with: Cursor
feat: add article template variant, npx CLI, and llms.txt auto-generation
- Template variants: `template: "paper"` (default) or `template: "article"`
in frontmatter. Article mode: centered single-column layout, no banner,
no figure numbering, no citation block (inspired by academic project pages)
- npx create-research-article: zero-dep CLI to scaffold new projects with
interactive prompts, template choice, and optional HF Space auto-deploy
- llms.txt: Astro integration that auto-generates /llms.txt at build time
with clean Markdown of the article content (llmstxt.org convention)
- Updated README, getting-started docs, and robots.txt
Made-with: Cursor
Start moon icon with opacity:0 and defer animated class via
requestAnimationFrame so the CSS transition only applies after
initial render. Also add continuous-glider data asset.
Made-with: Cursor
feat: improve PDF export with screenshot-based embed rendering and cover page layout
- Strip "Figure X:" prefixes from all embed desc props (auto-numbered now)
- Use set:html for desc rendering on dataviz page so <br>, <a>, <strong> etc. display correctly
Co-authored-by: Cursor <cursoragent@cursor.com>
docs: add dataviz page documentation to components chapter
Explains the /dataviz page feature: what it extracts (charts, images,
stacks, tables), how items are displayed, how to exclude items with
skipGallery, and how to batch-export charts as PNG/ZIP.
Co-authored-by: Cursor <cursoragent@cursor.com>
feat: backport dataviz page, embed export, and visual extraction from smol-training-playbook
- Add clear prerequisites section with NOTION_TOKEN and NOTION_PAGE_ID
- Add dedicated section explaining how to share page with integration
- Split usage into Option A (HF Space) and Option B (Local)
- Remove outdated pages.json references (now auto-generated)