Even if we just start with the assumption that n% of people will misunderstand, it still strikes me as unlikely that the best algorithm will be the current one.

Even if we just start with the assumption that n% of people will misunderstand, it still strikes me as unlikely that the best algorithm will be the current one. I think we have the next round of tests though. See how many people you need to randomly swap before Steffen is worse than back-to-front.