Honest Oort note: we’re tiny and tests are our best yield. Every prompt/project gets a 10‑case end‑to‑end test. It flags model drift, token quirks, provider breakage — and saves more time than another round of prompt polishing.