How this is made
The site is a SvelteKit app, prerendered to static files and served from the edge. Prose is written in Markdown (mdsvex), math is typeset with KaTeX at build time, and the charts are drawn with LayerCake — plain SVG, no charting library shipped to the browser. Styling is Tailwind; the type is set in a warm paper palette with a single accent per piece.
The look is deliberately closer to a printed section front than a dashboard: mono labels, a black italic headline, and one poster per story generated from the finding itself.
The data is git-scraped
Every number here comes from a committed snapshot, not a live API call at page load. A scheduled GitHub Action runs a small fetch script for each source — Supreme Court transcripts, Wikipedia edit histories, the NBA's own officiating reports, GitHub and Docker stats — and commits the normalized result back into the repository. The page reads that file.
The upside is that the whole site is a static, versioned artifact: the data has a history you can diff, a bad upstream day can't blank a chart, and nothing is hit at request time. Each data-driven story publishes the exact snapshot it reads, so any figure can be traced back to its source and checked.