Ask my notes.
This is my RAG That Doesn't Lie post, running live on itself: lexical retrieval over everything I've written here, answers that cite the exact section, and a hard gate that refuses before calling the model when retrieval confidence is below 4.5. You can watch it work — including the refusals.
Posts are chunked by section at build time (~60 chunks). Each question is ranked with hand-rolled BM25 — no vector database, because at this corpus size lexical retrieval is simpler, free, and explainable. If the top score clears 4.5, the top sections go to Claude with instructions to answer only from them and cite each claim. If it doesn't, you get an honest refusal instead of a confident guess. Every answer shows its retrieval so you can check my homework.