Originally Posted by JiruoVX
What problem? the BG3 chart looks completely fine with this 200 roll example. It almost hits the perfect average. Each group of 5 (1-5, 6-10, 11-15, and 16-20) are all near 50 rolls. This comparison doesn't highlight any problems with BG3s RNG.

Here is the problem. People think experienced randomness means you should get similar highs and low. This is true is large large samples. True randomness doesn't favor any one number over another. You rolling a 5 has no impact over what you will roll next. True randomness does not care about streaks or patterns. Any change to remove streaks is dice manipulation. In fact, we would expect streakiness in a truly random number generator.
I've highlighted the issue with BG3 (with "loaded dice" off) randomness. If you look at @Niara's plots, it is clear that BG3 rolls happen in a sine-wave pattern. Low rolls are preferentially followed by low rolls, etc.

I calculated the following sometime back using @Niara's dataset, but can't find the post so am typing it by memory:
rolls of 1-5 are followed by an average roll of ~9.8
rolls of 6-10 are followed by an average roll of ~10.3
rolls of 11-15 are followed by an average roll of ~10.8
rolls of 16-20 are followed by an average roll of ~11.3

It is possible that Niara encountered a weird patch of RNG, and that her data isn't representative of most rng in BG3. But the clear sine wave pattern is worrying, and I've seen no evidence against it (not many people have recorded the order of their rolls).

Originally Posted by JiruoVX
At this point, there isn't any concrete evidence pointing to their dice system is flawed. The real question is whether true randomness is something we as players really want in our video games. This very experience will always happen with true randomness.
You're incorrect here. I've collected 500 dice rolls (loaded dice = off) from players and the sample is inconsistent with a pure uniform sample.
n=508 rolls
average=11.16 (The expected standard deviation of a n=508 sample is 0.256, putting our average 2.6-sigma off from the expected average of 10.5)
chi^2=30.89 (greater than the critical value of 30.1, so we can reject the hypothesis that BG3's rng is generating an even distribution of numbers with 95% confidence)
Biggest Offenders
--1 appears 60% as often as it should, a 2.1-sigma difference from expectation
--6 appears 70% as often as it should, a 1.5-sigma difference
--17 appears 1.77x more often than it should, a 3.9-sigma difference
However, the "loaded dice" system (before the recent hotfix and using 750 rolls) was consistent with a uniform distribution.

Last edited by mrfuji3; 19/05/21 06:54 PM.