Colophon

Next.js and MDX, deployed on Vercel, with the posts kept as Markdown files in the repository rather than in a CMS. Code samples are highlighted at build time, so nothing has to boot a syntax highlighter in your browser to show you ten lines of YAML.

Diagrams are written as Mermaid inside the posts themselves and drawn on the server by Pele, in the surrounding typeface and colors, and nothing runs in your browser to draw them. The Markdown twin keeps them as Mermaid source, which anything that reads Mermaid can render. The post covers are illustrated by Gemini from a fixed prompt and a recurring cast, one per post, matched to what the piece is actually about rather than picked from stock.

Those covers and the favicon are the only parts of this site a model made. The posts are researched and argued over with one, never written by one, and every sentence here is a sentence I typed. The images admit it themselves. Each one carries the standard label for a machine-made picture inside the file, where any metadata reader will show it to you, and Google leaves an invisible mark in the pixels that its own tools can read back. The one thing missing is Google’s signed copy of that label, which does not survive an illustration being resized to fit this page.

Every page has a plain Markdown twin. Add .md to any post URL, or send an Accept: text/markdown header, and you get the source instead of the markup. There is an llms.txt mapping the site and an llms-full.txt carrying the full text of everything in one file. Some of that is for agents and some of it is for me: writing worth reading should be readable without a browser in the way, and a page that an agent can parse is usually a page that a screen reader can parse too.

Posts are files on disk rather than rows in a database, and the feed is generated when it is asked for. The only state the site keeps is the newsletter list and one like count per post, in Upstash. Contrast, focus states, and heading structure are held to WCAG AA, which is partly principle and partly the same observation as above: the structure that helps a screen reader is the structure that makes a page legible to anything else reading it. If something here is broken in your browser or your reader, I would rather hear about it than not.

What data the site collects is in the privacy notice.