How we work: sources, AI and editing

We use AI to work faster, not to replace checking. Here is exactly what happens before an article appears on this site.

1. Fresh primary sources

Topics come from official release feeds and project blogs. A source is only used if its page is reachable, has an explicit publication or update date within the last 30 days and contains enough readable text. Each article needs at least two such sources. Search snippets and summaries never count as evidence.

2. An AI-assisted draft

A language model writes a first draft from those sources and must cite them. Short news drafts are written by a model running on our own hardware; longer deep dives may use Google Gemini.

3. An independent review

A separate AI reviewer checks the draft against the saved sources: unsupported claims, wrong versions, broken code and missing citations send the draft back for at most two rounds of fixes, after which it is dropped. A second review checks the finished article and its illustration together.

4. A human decision

The editor, kord, reads every article before it can be published, and either approves that exact version, asks for specific changes or rejects it for good. Any change to the text, sources or illustration creates a new version that goes through review and approval again.

Illustrations

Cover images are generated by an image model on our own hardware for each article. They are illustrations, not photographs of real events, products or people, and they never contain brand logos.

What this does not guarantee

This process catches many problems, but not all of them. AI output can be wrong in ways that look convincing, and sources themselves can change or contain errors. That is why we link every source: please check details that matter to you, and tell us if something is wrong.