Colophon
How this site is built, and what it publishes for machines as well as people.
Most of what follows is invisible if you are just here to read. It is written down because this site tries to be a front door for people and machines, and the machine half is otherwise only discoverable by machines.
Built with
Ghost, self-hosted on my own server in a Docker stack behind Caddy, with Cloudflare in front. The theme is a fork of Neue, rebuilt in Tailwind. The writing is mine; where it is not, it says so.
Everything about how a piece was made is labelled — written by me, written with AI, or AI-authored. That is the AI provenance page, and the small mark on each post links to it.
For machines
The site publishes rather more machine-readable surface than it strictly needs to, partly because it is useful and partly because it is the thing I write about.
A curated index. llms.txt is a hand-written map of the site — who I am, the key pages, the writing worth starting with. llms-full.txt is the whole corpus in one file.
Every page as Markdown. Add .md to any post or page URL and you get the source instead of the HTML. So /loops-within-loops/ becomes /loops-within-loops.md. You can also ask for it by content type with an Accept: text/markdown header — same URL, no redirect.
An MCP server. machines.rafeblandford.com exposes my published career and writing to any agent that speaks the Model Context Protocol. Four tools — get_evidence, search_writing, get_post, check_experience — all read-only. There is deliberately no tool that contacts me: that stays a human decision.
It is worth knowing what that corpus is and is not. It contains what the site publishes, which is a selection. If it tells you a topic is not found, that means nothing here evidences it — not that I have no experience of it.
Skills you can install. An Agent Skill is a folder with a SKILL.md in it — plain instructions an agent reads to learn how to do something specific. There is one for using this site: which of the three interfaces above to reach for, and how to read what comes back without flattening the distinctions that matter. Resources has the install instructions, machines can read the index, and the source is on GitHub under Apache 2.0.
Discovery. The MCP endpoint is advertised four ways, because no single convention has won yet: a Link header on every response, a <link rel> in the HTML, an Agentmap: line in robots.txt, and an AI catalog. There is also an api-catalog following RFC 9727, which indexes the lot.
Usage terms. license.xml states them in RSL 1.0, and robots.txt carries a Content-Signal line. The short version: my own prose may be indexed, cited, and trained on. Images may not be trained on. That is not a preference but a rights question, since the image directory mixes my own covers with client logos and supplied photography I cannot license on anyone else's behalf.
These are preferences, expressed in standards that are young and lightly adopted. Publishing them does not mean anyone honours them, and none of it is a claim about enforceability.
AI crawlers are welcome. Explicitly, by name, in robots.txt — GPTBot, ClaudeBot, PerplexityBot, Google-Extended and the rest. That is a deliberate choice, not an oversight.
Feeds and contact
- RSS — everything I publish
- Email — occasional, and I do not track opens or clicks
- Security contact: security.txt
Privacy and analytics
No cookies. Analytics are self-hosted and cookieless, and store no IP address. The full detail, including what my server logs keep and for how long, is on the privacy page.
Corrections
If something here is wrong, tell me and I will fix it and say that I did.