This is experiment zero: get a technical notebook from an empty directory to a custom domain without dragging a client runtime along for the ride.

The question

How little machinery does a blog need if the interesting part is the experiments, not the publishing system?

The first cut uses Astro for typed content collections and MDX, then lets Cloudflare serve the generated assets at the edge. Pages arrive as HTML and CSS. There is no JavaScript in the default reading path.

The setup

Cloudflare’s framework initializer created the project and configured the Astro adapter:

npm create cloudflare@latest -- blog-r0 --framework=astro --platform=workers

The Worker is private to the custom domain: workers.dev and preview URLs are disabled. The public route is blog-alpha.abusybit.com.

Why MDX

Markdown is enough for prose and code. MDX leaves room for a future experiment to embed a tiny interactive trace, chart, or request inspector directly beside the explanation—without turning every page into an application.

Baseline

• One content collection with validated frontmatter
• Static article routes and an RSS feed
• Security headers delivered with the assets
• A custom 404 page
• Zero client-side JavaScript on the home page

That is enough infrastructure. The next entries can spend their complexity budget on the systems being measured.