RAG evaluation

Example entry. This is a tagged example so the wiki pipeline has something to render before the real Obsidian vault is connected. It is not a finished note.

Evaluating a retrieval-augmented system means measuring two things separately: whether retrieval found the right context, and whether generation used it well. A small golden dataset of questions with known-good answers makes both measurable instead of asserted.

This note links to hybrid-retrieval to show that wikilinks resolve.