matlab possibly?
Anyway I think it'd be worthwhile to do a set of simulations, not just the one simulation that yielded ~38k out of 10 mil. You could then draw a distribution of the results. Obviously, you would need to change the random seed each time to ensure you're not getting simulations that are identical to eachother.
It's not outside of the realm of possibility that you get a 38k when the mean (expected value) is 46k. In fact I imagine it'd be somewhere in the left tail of the distribution. Not impossible, but unlikely.
Edit: without going over in detail what others have said, is it possible that you're not counting 4 doubles in a row as 3 doubles, and so on? By the way it seems very odd that you need to set the previous roll to not be a double. Strictly speaking, 6 doubles in a row should count as 3 doubles happening twice, if you're just counting how many 3 doubles come up.
If this is for monopoly, swiggidy's proposal seems legit. Have your simulation roll 3 sets of die 10 million times, and count how many times they come up triple double.
You are correct in saying that the probability is significantly less than (1/6)^3 under these other assumptions.
Reply With Quote