What I particularly like is the fact that, from a purely data-driven perspective, the obvious choice is I-C; the rotary-mimicing layout. But (I conject) the lo

What I particularly like is the fact that, from a purely data-driven perspective, the obvious choice is I-C; the rotary-mimicing layout. But (I conject) the lower error rate and preference is merely down to familiarity of users with the older rotary system. 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. Which is not groundbreaking, but it's nice to have case studies.