The principle that increasing reasoning steps per agent improves both accuracy and efficiency independently of agent count.