One output cannot establish how well an AI system performs. Evaluate with multiple representative inputs, repeated runs, and confidence intervals.