From a statistical standpoint, 100 rolls could easily mean that it's just a too-small sample. If you roll that die 2 billion times and it's always 6, you have a problem. And really, saying "6 comes up 100 times in a row" is probably too extreme of an example. Let's say that if 6 comes up 50 times out of 100, that doesn't mean it's statistically significant when you're talking about random. Especially multi-factor random.
As in: let's say you have a single die that you roll for ten different purposes. Once for moving, once for what kind of card you get when you open a chest, and so on. Let's say that you happen to get a 6 each time you roll for movement for a while, but the die is really random on all the other purposes. Let's even say that those other 10 purposes don't even always happen every time you move, so there are 0-n extra rolls of the die between each movement roll. So what does that tell you? Absolutely nothing. That's actually what's going on here in AI War.
At any rate, complain about the RNG all you like, but I've looked into it very recently and it's working fine. The things that I thought were an issue recently boiled down into three problems:
1. The fact that we were reusing the same seed all the time due to a recent bug outside the RNG itself, which we fixed.
2. The fact that the RNG was very unlikely to choose the top-most item out of a list of items, because of our use of FInt (which we fixed). This has to do with how we were "shifting" the actual UInt random value to a specific range in the integer or floating point space.
3. Some other things we'd been worried about that actually turned out not to be problems once we fixed #1 and #2. They were false flags, downstream issues based on the above.
As Bob alludes to, essentially there will always be complaints about any RNG, in any game, because statistical clumping is real and happens frequently. It's often-discussed in game design circles. A number of game designers actually take steps to make sure that "random" in their cases winds up being "distributed at a certain ratio," even though that's not really random, but just because that's what players expect.
In the case of AI War, there are many cases where we use weighted randoms to set the likelihood of certain ships in the hands of the AI, etc. But even then we're just affecting the composition of the source list, not the actual RNG itself. Thus even if something is semantically "1% likely to happen" based on its frequency in the list, it could still be valid to show up a lot more frequently. Especially because of the multi-factor stuff I mentioned above.
