Every game ends up having discussions like these...

*A) I've tried X 1000 times and only gotten y result 3 times, it's borked!*

B) Dude, your sample is too small. If you do it 100,000 times it'll be good.

A) and B) argue awhile with others chiming in how their anecdotes prove A) or B) is right/wrong/crazy.

C) *I've* done (insert OCD-level insane number of reps) and only gotten y __ times. A) is right, something is borked.

Well, let me toss something out onto the fire.

Lets say a given RNG needs a sample set of at least 100,000 to approach a truly random distribution. The problem is, to see that even distribution, every result of that RNG needs to be in the data set. Which is just fine if you're feeding it one operation and looking at the results of 100,000 consecutive iterations.

Unfortunately, in a game, that RNG is working on hundreds of different data sets at any given time.

I really don't see how you could ever collect enough data to expect truly random results. The game would have to be coded so that each specific function had its own dedicated RNG associated with it to approach any kind of even distribution.