The crux of this whole problem is that most humans are terrible at recognizing randomness. Humans expect that randomness implies some kind of uniformity free from streaks. It does not.

If you want to check for fair dice, there is an ideal statistical test for that. The Chi Square test. It is usually covered in every intro statistics class, and they use dice as an example because everyone thinks they understand them.

I coded an NWN script to roll 2000 D20s and accumulate the results. This is near the maximum because the scripting engine only allows a set number of instructions to each script for performance reasons and to break endless loops.

I output the accumulators, calculate the overall average D20 value, and the Chi Square statistic for the run. I ran it 10 times, So 20000 D20 rolls overall.

At P = .05, the Chi Square Critical Value (for 19 degrees of freedom) is 30.144.

To disprove the Null hypothesis (The NWN Dice are fair), we need a Chi Square Statistic >= 30.144.

At no time in 10 runs was it even close. There is no statistical basis for claiming the NWN D20 generator is unfair.

Attached is the output of my last run. Chi Square was typically between 10 and 20, and the highest was 22.