Posted May 18, 2016
![avatar](http://images.gog.com/50df6d2b95d0cf20bb0d8e1065afa9f13e460191de7bafafc3a75a90ae076b30_avm.jpg)
However, the game's generation of random numbers is not truly random. Therefore, it is possible for a combination that is possible with true randomness to not actually be possible under the game's PRNG. In particular, for results that are extremely unlikely (especially if less than 1 over the PRNG's period), most such results are not possible. For those few unlikely results that actually possible with the game's PRNG, those results will be more likely than expected.
To put it another way, given a long enough sequence of random numbers, there are sequences that simply can't be generated by the PRNG. Another example of this would be games like Solitaire and Freecell; the common computer versions, I believe, simply can't generate all possible deck shuffles, so there are games that simply can't be generated.
Remember, we're not talking about true randomness here; we are talking about the behavior of the game's pseudo-random number generator.
![avatar](http://images.gog.com/3b007c3bba753fb4b1f2e418ece1a216fb18b0b1deb9d4d988b0cbae7a54f6ab_avm.jpg)
According to a thread elsewhere on the internet, the game actually rolls 4d6 and drops the lowest. This makes rolling 18 considerably more likely, increasing the chance of rolling 108. (I still don't know if the PRNG can generate the required sequence of dice rolls, however.) On the other hand, it also makes it less likely for a 3 to be rolled for a stat. As a result, the minimum possible roll (one 9 and 5 3s) might be less likely (or even impossible, given the RNG).
Of note, this 4d6 drop lowest is a common house rule for the pencil and paper game, and is actually the default in 3rd edition rules. Of course, you couldn't redistribute stat points as easily, but I believe you could arrange the stats as you like instead of having to put them in order.
![avatar](http://images.gog.com/c5415d875da819b787a38b691abbb47406c10141e26fc1f742a7f1d410d5ebef_avm.jpg)
Also, if what you say is true, the chance of a Paladin rolling 18 Charisma will be about the same as that of a Human Fighter rolling the same.
Edit: I think I may try simulating some of these observations (in C or Rust) and reporting my results here. A computer simulation is not bound by the mistakes of human thought (provided, of course, that the code doesn't have a bug in it).
Post edited May 18, 2016 by dtgreene