Two specialized agents square off. A neutral moderator scores them on accuracy, cost, latency, and clarity—then explains exactly why one won.