Knowledge base checks itself, no language model

Request: „End of day — save, check, update the daily note.“

1. Every note carries provenance, last check and review date
2. Deterministic check scripts control links, routing, metadata and regions — without a language model, with an unambiguous result
3. Every day a quality gate answers 50 automatically generated domain questions against the brain and scores structure and answer quality
4. At the end of the day: save open files, update daily note and tasks, run the checks, secure the state in a personal branch

In any organisation, knowledge bases age quietly; dead links and missing provenance surface only when someone needs the note.

Assess the findings; content is not shortened without asking.

While “optimising” an overview page the AI shortened content — everything was restored verbatim and verified by comparing a good 70 commands; since then: no shortening without asking.

Any organisation with a knowledge base that AI accesses. An AI that “optimises” likes to shorten — which is why code checks, not a model.

An AI is only as good as the knowledge it accesses — and knowledge ages quietly. So every note carries its provenance and check date; around 1,650 existing notes across eight brains were back-filled for it. Check scripts without a language model control links, routing and structure every day, and a quality gate asks the brain 50 domain questions daily. At the end of the day the AI tidies up and secures the state in a personal branch. One lesson came from a mistake: while “optimising” an overview page the AI shortened content — everything was restored verbatim, and shortening without asking has been off limits since.