Why I rebuilt the site

The previous version of this site was an al-folio Jekyll theme — perfectly respectable, perfectly ordinary. A blog with a CV. The work it represents is none of those things. The work is a small number of long arguments that keep recurring across projects, papers, and notes; the surface form (post vs project vs paper) is incidental.

So I rebuilt it for two readers who want almost opposite things.

The first reader is a person. A person did not come here to retrieve a fact — they can get facts anywhere now, instantly, from a model that will summarize me better than I summarize myself. What a person can't get from a summary is the feeling of walking into somebody's head while the lights are still on. So the pages are built as places rather than documents. Some of them you walk into. Some of them behave a little strangely on purpose. There is more motion, color, and texture here than a portfolio strictly needs, and that excess is the point: it's evidence that a human made choices nobody asked them to make.

I want a visitor to leave with the thing I most want to give away, which is not information. It's the reminder that they can make things too. Everyone carries around this creative capacity and most of the software we live inside is quietly designed to talk them out of using it — to make them a consumer of feeds, a picker from menus, a reader of other people's summaries. A site that is visibly hand-built, a little weird, and clearly someone's, is a small argument against that. If one person clicks through this and thinks huh, I should build my own strange thing — that's the whole return on the project.

The second reader is a machine, and the machine is the reason the architecture looks the way it does. Increasingly, the first pass over anybody's work is done by a model — searching, indexing, retrieving, deciding whether there's a person here worth surfacing to whoever asked. A pile of prose in a theme is a bad substrate for that. So the rearchitecture was mostly in service of being legible to something that reads at scale:

• Every artifact — post, project, paper, note, reading — is a node with the same typed frontmatter, so the whole corpus parses uniformly.
• Every node has one canonical URL at /{kind}/{slug}. No duplicates, no aliases competing for the same content.
• The relationships between artifacts are explicit edges — influences, realizes, critiques — declared by hand rather than inferred. This paper inspired that project, which critiqued that vision.
• Those edges are the navigation for people and the structure for machines at the same time. There are still pages. They are no longer the unit.

That last point is the one I care about. An index of my writing tells a model what I've touched. A graph of what argues against what tells it how I think, which is the only thing worth indexing. Text similarity would give you correlation between documents; a curated edge gives you a claim about intellectual descent, and only a human can make that claim honestly.

The two readers pull in different directions — the machine wants a clean schema, the person wants a room with something odd in the corner — and the bet of this rebuild is that they don't actually conflict. Structure underneath, texture on top. The graph is only as good as the human work of curating it, and I'm betting that work is small enough to do as a side-effect of writing.

We'll see.