Goal and sources first. Then run sample extract, or let an agent call extract_claims. This page never scrapes.