How these articles are made

Stated plainly, because you should be able to judge the process and not just the prose.

Research and drafting

Articles begin from primary sources: specifications, model cards, papers and documentation, rather than from secondary coverage. AI tooling is used for research synthesis and first drafts. Nothing publishes on that basis alone.

How an article reaches this site

Articles are drafted and published by an automated daily pipeline, and they go live without a human reading them first. That is an unusual thing to admit, so here is exactly what stands in the way of a bad article instead.

Every draft must pass a fixed set of automated gates before the build will produce a page at all: a prose linter, a content validator that enforces the sourcing and structure rules below, link and schema checks, and a figure and section audit. A failed gate blocks publication rather than raising a warning. The draft is also required to open every source URL and confirm it supports the claim attached to it.

Human review happens after publication rather than before it, and it is mine. I read what the pipeline ships, and when something is wrong I correct it in place and date the correction in public. The review is real and the accountability is real; the sequence is the part that differs from how most publications describe themselves, and pretending otherwise would make this page the least trustworthy thing on the site.

Sourcing standard

• Minimum two primary sources per article, listed with publisher and date.
• Primary means the specification, filing, regulation or paper itself. A specification is cited from its own text, such as the Model Context Protocol specification, and a regulation from the issuing body, such as the European Commission on the AI Act. Coverage of those documents is not a substitute for them.
• Vendor claims are attributed as vendor claims, never stated as findings.
• Where a number is measured, the method is described so it can be re-run.
• Topic hub pages carry the same burden as articles. Each one lists its own sources, because a page that asserts things should show its working whether or not it is called an article.

Updates and corrections

Evergreen articles are reviewed on a schedule. Substantive changes are dated in a visible revision history on the article and reflected in its machine-readable metadata. Corrections are never made silently.

Machine access

Content is deliberately open to AI crawlers and agents. Every article is available as markdown by appending .md to its URL, and the full index is published at posts.json and llms.txt. Crawler policy is stated in robots.txt under RFC 9309, alongside Cloudflare's Content Signals. Every page also carries a schema.org entity graph with its citations attached. If a model cites this site, we would like the citation to be accurate, which is easier when the source text is easy to read and the references travel with it.