I wondered if someone would walk into this one.

Imagine you have ten dice, all of which have different numbers on. You roll all ten repeatedly, sum the numbers, and do some statistics on the results. You can calculate the mean and dispersion, and these will give you accurate probabilities for future rolls.

If you extract one die and start it rolling it on its own, you can make no assumptions about its behaviour from the previous results, and the results you get from one die tell you nothing much about the group roll results. Because each die is different.

Reasenberg is in effect looking at the stats for a whole bucket of dice. Martin is saying, "hey! I've got two dice here and they're different from that!".