A guide for readers that do not scroll

Why another file

robots.txt and sitemap.xml tell search engines what can be indexed. But an LLM assistant reading a page or answering a question about the atlas is not indexing HTML - it parses the page live and re-guesses the site’s structure every time. The unofficial llmstxt.org convention offers a shorter path: one Markdown file at the domain root that states plainly what the site is and where to go on it.

What is in it

/llms.txt: the atlas’s name and description - the same one t(locale, 'description') gives - followed immediately by the alpha status, stated honestly: most records have not been reviewed by a person yet, and the “sources gathered automatically” badge on every record page says so, not only here. Then links to the axis, the full card list, the methodology, the source bibliography, the tags, the people/organizations/models indexes, the graph, and this very blog. The last line is the sitemap and a note that the same routes also exist under /en/.

What was checked

The file sits outside the site’s HTML pages: validate:public keeps only the built pages and the sitemap consistent with each other, so a new static file does not touch that contract and needs no gate of its own. .txt is already a known extension for the local preview server - the same one that serves robots.txt. npm run verify:code passed in full, with no change to the byte budgets.