Documentation that checks itself

About this project

Documentation rots quietly. A page says a service has eight modules, the code has thirteen, and nobody notices until someone new trusts the page. We got tired of that failing silently, so we made it measurable.

The tool reads three things at once: what the code actually contains, the notes we write alongside it, and the documentation as published. It maps how they connect, records what every claim is based on, and reports what no longer lines up. Change a module and you can ask which pages just became wrong.

It never writes prose. It only compares and counts, which is exactly what makes it worth trusting, and why our own AI tooling asks it instead of guessing at the codebase. Wording and translation sit on top, where being wrong is cheap to fix.

Highlights

• Compares the code, our notes, and the published docs in one pass
• Records what every claim is based on, so nothing rests on memory
• Reports the pages that no longer match the code
• Change a module, see which pages just went stale
• It compares and counts, never writes, so tooling can build on it safely