Static Page Stack (SOTA, 2026)
Self-Evolving Skill: This skill improves through use. If instructions are wrong, parameters drifted, or a workaround was needed — fix this file immediately, don't defer. Only update for real, reproducible issues.
When to use this skill
The user wants to produce static HTML pages that combine three aesthetic DNAs:
- Magazine / data-journalism (The Pudding, NYT Upshot)
- Technical documentation (Material/Docusaurus/Starlight)
- Computational notebook (Quarto/Observable/Jupyter Book)
…and embed three kinds of SVG:
- Programmatic charts (Vega-Lite / Observable Plot / D3)
- Hand-authored inline SVG (illustrative)
- Whiteboard-style (Excalidraw / tldraw)
Triggers include: "create a static page", "build a dashboard", "design a post-mortem", "set up a technical-doc site", "what HTML/CSS stack should I use", "this Tailwind soup is unmaintainable".
The recommended stack (TL;DR)
| Layer | Pick | Backup | Why |
|---|---|---|---|
| Authoring | Quarto | Quartz, Jupyter Book 2 / MyST | Multi-page, native code+prose+math+execution, MIT, used by Bundesbank/BIS/USGS |
| Visual system | Open Props + native modern CSS | Pico.css | Replaces Tailwind utility soup with semantic tokens (4 KB Brotli); OKLCH + container queries + light-dark() are production-ready in 2026 |
| Typography | Inter + Crimson Pro + Fira Code | IBM Plex family | All FOSS; pairing tested at NYT/Pudding-grade |
| Programmatic charts | Observable Plot | Vega-Lite, ECharts | Cleaner defaults than Vega-Lite, ~95% of needs without going to D3, ISC, Bostock-led |
| Whiteboard | Excalidraw + excalidraw_export CLI |
tldraw (note: $6k/yr commercial license) | MIT, mature, build-time SVG render works |
| Hand-drawn programmatic | rough.js (sparingly) | roughViz, chart.xkcd | Only for "this is exploratory / uncertainty implied" framing |
| Hand-authored inline SVG | CSS-vars-in-SVG pattern | <symbol><use> for shared icons |
var(--token) propagates through inline SVG, giving single design-token surface |
| Scrollytelling | Closeread (Quarto-native) | Scrollama | Closeread is purpose-built for Quarto; Scrollama (~7 KB) for non-Quarto |
Critical clarifications (don't skip these)
"XHTML" in 2026 means semantic HTML5, not strict XHTML
Nobody writes literal application/xhtml+xml any more. What "elegant XHTML" actually means in modern usage is well-formed semantic HTML5 enforced by the build tool. Quarto outputs exactly this — <article>, <aside>, <figure> — without making you hand-author every angle bracket. If a user says "XHTML", clarify they probably mean semantic HTML5 with structural discipline.
Tailwind itself isn't the problem
The problem is Tailwind CDN + ad-hoc utility-class soup, no design system, regenerated each time. Open Props gives you the same "drop-in tokens for everything" feel but as semantic CSS variables that propagate INTO inline SVG. That last property is what makes the stack coherent.
Distill is dead, Idyll is dormant — don't recommend them
- Distill.pub stopped publishing in 2021. The
distillpub/templaterepo looks alive (972 stars) but has no funding story. - Idyll-lang's last meaningful update was Feb 2023.
- The 2026 spiritual successor is Quarto's article layout + Closeread, not a Distill fork.
tldraw has a non-FOSS commercial clause
tldraw 2.x+ is source-available under a dual license: hobby (watermarked, non-commercial) or $6,000/year for commercial. Default to Excalidraw (MIT) unless tldraw's specific features (interactive embeds, polished hand-drawn aesthetic) justify the cost.
Decision flow
User wants a static page →
├─ Single document, code-heavy, math? → Quarto (single-file render)
├─ Multi-page investigation/dashboard? → Quarto with sidebar nav OR Quartz (digital garden) if cross-linking dominates
├─ "Pure data dashboard, no narrative"? → Observable Framework (caveat: Observable Inc layoff risk)
├─ "I want sidebar nav, search, code blocks"? → Astro Starlight + MDX (functional docs aesthetic)
└─ "I want maximum flexibility, willing to design"? → Astro + custom theme + Open Props
User wants a chart →
├─ Static, magazine-quality default? → Observable Plot (default pick)
├─ Need interactivity, dynamic spec generation? → Vega-Lite
├─ Need polished interactive dashboards, Canvas OK? → ECharts
├─ Want sketchy "this is exploratory" framing? → rough.js / roughViz
└─ Need bespoke custom path/glyph? → D3 (rare; Plot covers ~95%)
User wants a diagram →
├─ Architecture, flow, illustrative? → Excalidraw + excalidraw_export
├─ Premium aesthetic, budget for $6k/yr? → tldraw + tldraw-cli
├─ Inline SVG with design tokens? → Hand-author with var(--token-*)
└─ Reused across many pages? → <symbol><use> pattern (caveat: shadow DOM blocks per-instance CSS)
Concrete migration: Tailwind CDN soup → Quarto + Open Props
Phased migration (~5-6 weeks total for a multi-spoke dashboard):
Phase 1 — Pilot (1-2 weeks): Convert one page/spoke to Quarto. Match current visual output by writing custom SCSS theme that reuses existing palette but expressed as Open Props tokens + OKLCH. Ship as separate dashboard-quarto/ directory.
Phase 2 — Aesthetic upgrade (1 week): Swap to Inter + Crimson Pro + Fira Code, replace ad-hoc Tailwind colors with OKLCH-generated palette ramps from oklch.fyi, standardize figure/chart components.
Phase 3 — Charts (1 week): Convert hand-authored data-bearing SVGs to Observable Plot specs. Illustrative SVGs (architecture, flowcharts) move to Excalidraw.
Phase 4 — Scrollytelling (1 week): Pick 2-3 mechanism explainer pages. Rewrite as Closeread scroll-driven explainers.
Phase 5 — Roll out (~1 week per remaining page): Use phase 1's templates.
Anti-patterns to call out
When you see these in user code, recommend alternatives from the stack:
<script src="https://cdn.tailwindcss.com">in a static page → Use Quarto + Open Props instead. Tailwind CDN doesn't tree-shake → bloated CSS, no design system, every regeneration drifts.- Inline base64-encoded SVG
<img>→ Use inline<svg>withvar(--token)for design-system propagation, OR<svg><use href="#sym">for shared icons. - Hand-typed numeric values in charts ("forecast: 189K, actual: 223K") → Use Observable Plot reading from a CSV/parquet so the chart re-renders if the number is corrected.
style="color: #abc123"repeated across pages → Define an OKLCH palette in:root, expose via Open Props tokens, reference viavar(--color-*).- Mermaid/Graphviz in a project that wants editorial-quality diagrams → Switch to Excalidraw or hand-authored inline SVG. (Mermaid is great for "boxes-and-arrows generated from text" but never reaches editorial quality.)
Honest tradeoffs (don't oversell)
- Quarto's default theme is not magazine-grade. Bundesbank-quality output exists, but you'll spend ~1-2 weeks on a custom SCSS theme. If the user wants plug-and-play magazine aesthetic with zero design work, no FOSS option in 2026 truly delivers — The Pudding's
svelte-starteris closest but it's Svelte+Vite (not Quarto). - Observable Framework has post-layoff risk at Observable Inc (2023, possibly 2025). ISC license means a fork is viable but corporate backing is thinner than Posit's commitment to Quarto.
- Open Props requires learning the token naming scheme. Not "plug and play" like Tailwind. The payoff is a coherent design system, but the first week is steeper.
How to apply this skill
- Identify the user's actual ask — single page or multi-page? Code-heavy or prose-heavy? Charts or just diagrams? Pure narrative or data-driven?
- Use the decision flow above to pick the framework + chart-lib + diagram-lib trio.
- Default to Quarto + Open Props + Observable Plot + Excalidraw unless something specific rules it out.
- Reference
references/full-research-synthesis.mdfor the per-layer evidence summary if the user wants to see the comparative analysis. - Don't recommend dead/dormant tools (Distill, Idyll, plain Tailwind CDN).
- Pull starter templates from the URLs in
references/sota-urls.mdwhen scaffolding a new project.
Post-Execution Reflection
If a recommendation didn't fit the user's needs in practice (e.g., Quarto's build was too heavy for their scale, or Observable Plot couldn't express their viz), update this skill — note the case and the alternative that actually worked. Don't defer to "next time."