Use when turning a rough ML idea into a runnable experiment - define the experiment card (question, hypothesis, baseline, primary metric, budget, cheapest verification rung) before writing training code.
