Comparing two generated outputs against each other to determine which better satisfies requirements, rather than scoring each independently.