The problem with the loaded dice solution is then that it isn't a solution -- it is a crutch. The difference between a solution and a crutch is that with a solution the problem is solved, but with a crutch, the problem is worked around but remains. The RNG is like a leaking pipe and the loaded die crutch is a bucket placed under it instead of fixing the pipe. If they said that hey this is a short time measure because rewriting the RNG is a big procedure and will take time, that would be fine. But if this is their long time "solution"... well, it doesn't solve anything.
Wouldn't "solving RNG" mean changing the system? If the system is "roll 20", there is nothing you can do to improve it outside changing the rules or loading the dice. Or you can cheat without telling the player as many games do - not allowing to roll low many times in the row, forcing a good roll if it didn't happen in a while etc. I would rather get a system which works in the way it tells you to do, buy I won't argue that just rolling d20 is not a good system. It ain't good here, it ain't good in Kingmaker.
No. A RNG isn't like a die. It's not actually random. There's no way to just tell code to "roll a d20" or "draw a random number". The code is purely logical, it doesn't know what "random" means. So the coder needs to build a system to tell the program what random is and how to draw a "random" number. This is done through matemathical formulae which attempts to simulate randomness.
So RNGs arent truly random -- they're approximations of randomness. And differently coded RNGs will simulate randomness in different ways (with different math formulae).
Mathematically speaking if you roll a d20 an infinite amount of times each result should have a 1/20 chance of appearing. This is what the algorithms try to mimic. In the case of Larian's RNG, testing has shown that with a very large number of rolls, it does start to approximate this result. However, while doing this, each independent roll shows a likelihood of being similar to previous rolls -- something that you don't want from an RNG. This is what Niara meant by "appearing in a sine wave" and I by "clustered results". It means that youare going to see rows of similar rolls instead of random numbers.
This likely happened because when they wrote the RNG they only checked the long term result -- run the code a thousand times and see how many times each number comes up, and if it approximates to 1/20 to each number call it a success. But you also need to examine the spread of each number throughout the 1000 rolls to make sure they don't cluster together in a way that a good RNG shouldn't.
Edit: ninja'd by Ninjara while I wrote
