Colophon

How this site is built, and — more interestingly — what stops it from saying something it shouldn't.

This is a static site. Every page is HTML written to disk ahead of time and handed straight to your browser. There is no database behind it, no server deciding what to show you, and no JavaScript at all — not a minimal amount, none. If you turn scripting off entirely, this page is exactly what it was.

That last part is not minimalism for its own sake. Every script on a page is something that can misbehave, and every third-party script is a decision to trust someone else's server on your behalf. Skipping them is the cheapest security control I know: nothing to keep patched, nothing to audit, nothing to leak. So there are no analytics, no fonts loaded from somewhere else, no embedded widgets. Nobody is counting you.

The parts

It is built with Astro, which turns a folder of Markdown files and templates into plain HTML, and served by Cloudflare. One page layout. One stylesheet. Adding a page here means adding a file — no list to update, no menu to edit. The navigation you see at the top is worked out at build time from what actually exists, which is why it cannot drift out of date or link to something that isn't there.

A section with nothing in it doesn't appear at all. I would rather a link not exist than lead you to a page saying "coming soon".

The guardrails

Every page arrives with a strict content security policy — a short list, sent with the page, telling your browser what it is allowed to load. This one says: my own stylesheet, and nothing else. If something ever managed to inject a script into a page, the browser would refuse to run it. The policy is deliberately tight enough that the easy workarounds are not available to me either, which is the point of writing it down.

There is also a test suite. It builds the whole site and then reads the finished HTML looking for things that must never appear on it — the sort of detail that is fine in a private notebook and wrong on a public page. It checks what actually shipped, not what I meant to write, because those are different things and only one of them is served.

And because a test that has never failed proves nothing, I break it on purpose. Each run deliberately plants one of the forbidden things in the page, rebuilds, and confirms the suite catches it — then puts everything back. A check that has never gone red is decoration. This one has to earn its keep every time it runs.

Separately, an audit runs against the live site from outside, the way you reach it, and grades the headers it actually gets back. Configuration files describe intent; only a request to the real thing tells you what is true.

Who wrote it

Mostly AI agents, working to a written plan, under the tests and checks above. That is the interesting part to me: not that a model can produce a web page — it can — but that the interesting engineering is in building the constraints it works inside. The tests, the boundaries, the thing that refuses to publish. I direct the work and hold the line on what is allowed to ship. The typing is increasingly not the job.

This page is the smallest example of that. It exists because I added one file and one line to a list. Everything else — its place in the menu, its address, its inclusion in the checks — followed on its own.