What I particularly like is the fact that, from a purely data-driven perspective, the obvious choice is I-C; the rotary-mimicing layout.<p>But (I conject) the lower error rate and preference is merely down to familiarity of users with the older rotary system.<p>So the lesson, for me anyway, is that listening to your users or to your data without applying your own reasoning can easily lead you to the wrong conclusion.<p>Which is not groundbreaking, but it's nice to have case studies.