Pulls out the real goal behind a request rather than the request itself, by running a one-question-at-a-time interview until you can predict the user's answers. Use when a request lacks who/why/what-success ("build X" with no audience or motivation), when the user invokes it directly ("interview me", "grill me", "stress-test this", "are we sure?"), or when you notice yourself quietly inventing requirements before a plan, spec, or code exists.
