We could also think about the first model just having three different permutations of orders, and multiply it by 3 to get the same result. Overall, this doesn't change the posterior prediction very ...