Grounding

Grounding aims to make a model use trusted context supplied with the request instead of relying only on broad training knowledge. In KobiGPT that context is department-scoped document passages.

The connection improves auditability. A user can compare the answer with the passage it cites; fluent model prose is not sufficient evidence by itself.

If a document from the wrong company or department is supplied, an answer may look sourced while relying on the wrong context. Tenant, assistant, and user scope are prerequisites.

The role model controls which users can reach which assistants. Even with correct access, freshness and department assignment still need review.

Grounding does not automatically resolve contradictions and cannot create information absent from the archive. If two policy versions are in scope, ambiguity can remain.

For consequential processes, read the source passage and have an authorised person approve the result. Grounding supports review; it does not remove human oversight.

A useful way to evaluate grounding is to follow one real question from the source document to the final answer. Record which file was selected, what context reached the model, and what a reviewer would need to verify. This turns a definition into an operational check and makes the result comparable across teams.

The same check should include ownership and change management. Decide who updates the relevant documents, how an outdated result is reported, and which access boundary applies. KobiGPT can provide the assistant and the cited document context, but the organisation still owns the source material, permissions, and the decision made from the answer.