Larian Studios Forums Baldur's Gate III Baldur's Gate III - Suggestions & Feedback please the truth, dice result really random

Niara is a bit statistically tuckered out at the moment, and is focusing her efforts on other things. I'm keeping on eye on the discussion, and I can field any questions that pop up if I'm able to, but I'm not currently up to gathering more samples, over other uses of my energy. Still it sounds like several others have taken it upon themselves to make recordings as well, so that's good.

If you folks are looking at the flow of the RNG make sure to preserve in your your records the order in which things are recorded as well, and where strings of results were broken by other events if they are - such as where your play sessions ended each time as you gathered data.

I am tracking them in order and did note any test/experiment segments. The one thing I really want to do better is tracking enemy rolls. I didn't track them for some battles. But it should work out after doing all routes for the goblin camp.

Playing again last night my MC tried to lift the burning wood off the guy in the burning Manor House (I forget his name, but I am sure you know the one). I had to make 9 on a d20, failed the first try, and admit just reloaded a save. I am sometimes tempted to do this, but usually resist. However, this time, out of bloody mindedness, I decided to keep reloading until I succeeded. It took me eight attempts (i.e. I failed 7 times).

The probability of this happening, getting a fail 7 times in a row, is (8/20)^7 = 0.0016384 or about 1 in 610. Hmmm

The probability of it happening right at this point maybe be fairly low, but this is just when you happened to notice it. The probability of having 7 low rolls in a row at some point is actually fairly good. It would be weird if it didn't happen from time to time.

Alright, the Data is in.


First and foremost, Overall there were definitely less streaks than gathered data before weighted dice. While documenting combat rolls Advantage and Disadvantage started to centralize very early for the player and enemies in combat. They both match what we would expect from Advantage and Disadvantage. Normal attack rolls for the player and enemy had an interesting story.

The Data

Normal attack rolls (Player)
Sample size (508)
Average (25.40)
Standard Deviation (4.37)

Advantage & Disadvantage (Player)
Sample: Advantage (235) Disadvantage (217)

When tracking true random, I only had one number exceed two standard deviations and that was a smaller sample, this time both six and ten are beyond two standard deviations.
Frequencies of 17-34 would be normal and the sample had 16 sixes and 35 tens. Maybe as the sample size grows these might fall into normal, but each encounter it stood out that 6's were rare and 10's were common.

This is in contrast to Normal Enemy Rolls, which below are shown to all be within the expected range. With that sample's standard deviation, frequencies of 3-12 would be normal. No roll in the enemy sample has a frequency above or below those thresholds.

So, my conclusion on weighted dice is that yes they did reduce streaks, and the player might have a reduced chance to roll a six with an increased chance of a ten. I'm really looking forward to seeing what other samples show. Did other players experience the same?


Normal attack rolls (Enemy)
Sample size (146)
Average (7.30)
Standard Deviation (2.43)

Advantage & Disadvantage (Enemy)
Sample: Advantage (41) Disadvantage (23)

I think the hardest fight for me was the Red Caps when I pissed off Ethel at lv3. Not Bulette, not Githyanki patrol who I trampled at lv4.

Those damned Red caps were one shotting my characters through mirror images, but luckily of the 4, 1 bugged out and kept standing still while the caster just decided to never come and just chill in the swamp. So I lived.

The probability of rolling on a die does not change because you "notice it". The probably of rolling a 20 sided die 7 times and getting a number 8 or lower, is exactly what the poster said.

Larian's RNG code is demonstrably broken.

Well with how many people have played Baldur's Gate 3, 1 in 610 is very possible. That's what makes it random, it can randomly happen and subvert your expectations.

I have my numbers too but I don't know how to dataize them...

Yeah, a 1 in 610 chance boils down to that for every 610 people who play the game, you're likely to get one who experiences these numbers. Given that it's been sold by millions and probably played more than once by a fair few of them such runs are definitely expected.

Even with something like DragonSnooz did, meticulously and scientifically record the rolls in a playthrough, you might even see a run with as high a chance as one in a million. Because that playthrough is in fact one playthrough of millions made.

Humans also tend to look for patterns, particularly ones that stand out good or bad, like a series of good runs or bad. I've no idea about the RNG system, but I would have thought that in a modern game that relies on RNG this would be one component they made sure they got right.

That said, there is a difference in how it performs and how we as players experience it. When I see a target of 50%, I assume it's going to miss more than hit, usually at a rate of about 3 to one. And it generally feels like it does so. I completely acknowledge this might be my bias, but it definitely feels askew to me. I am prepared to say that without the actual figures to know if this is right, this is entirely a subjective opinion, but I definitely have it.

We do know that it really doesn't get the whole target/result thing right when you use something like guidance or other variable assists..

Once with my ranger, advantage, bless and a +1 bow I was told I had a 95% chance to hit. I missed. Intuitively you would think I rolled critical misses. Of course it could mean that I had a really low target to hit and I missed both rolls. I looked at the combat rolls and it turned out I had rolled a 4 as my best roll. Turns out I needed to hit a 5 and didn't.

Now, I'm not a maths person, so maybe someone who is could help me. How does rolling a 5 (a 1/4 chance to miss) equate to a 95% chance to succeed? Even with 2 dice rolls. My only way of understanding it is that when it came to the calculations of how likely it was to hit, it rolled high on both bless rolls. Then when it came to actually make the roll for combat, it rolled low on both these.

I really think when it come to bless or guidance or any other numerical variable that modifies the dice roll, it should roll once when you estimate the target you need, then apply that result to the actual roll, rather than re-rolling. Otherwise it makes a mockery of the whole target roll idea and leads to bizarre situations where you roll higher than you need, yet still fail.

Ok,
here is the thing with programmed randomisation. It is not possible. What we have in Games is as somewhere mentioned above, a pseudo-random number generator. Computers calculate the numbers and you cannot calculate true randomness. So we get something that is nearly random. It is like those calculations you (hopefully) had in school with the funny lim->0 or lim->infinity.
That's a bit of an easy explanation but should suffice here for mere understanding.

Those RNGs tend to get streaky sometimes. I guess everyone playing games has expirienced this at some point. Fun fact is that with true randomness even that would not be impossible. So it usually works out. Streaks are very annoying when playing though. I guess that is why they created that loaded dice option. I have not looked into that but i think they flatten out those streaks with it somehow. That again goes farther away from RNG.

That is my knowledge so far. If there is some math student around they may have a better explanation on propability calculation than i have.

Omigod! Who are you? I'll tell you what this reminds me of. I heard this story about a statistician who got thrown into prison. And he kept track of the days by keeping a tally chart on his cell wall. If I read between the lines, it seemed like he was paranoid that if he didn't do so, that the prison would be able to convince him that he hadn't been there for as long as he knew he'd been there.

Rolling a 5 normally is a 16/20 chance = 80% (not 3/4).
Rolling a 5 with advantage is a 1-(4/20)^2 = 96% chance. (calculated as "100% minus the chance that neither dice rolls higher than a 4")

Seems fine to me.
Edit: I think Larian should list the number on the d20 you need to roll, not the percentage chance of hitting. For this exact reason: it feels more unfair when you miss with 95%, than you miss on "Need to roll a 5 with advantage"

But yeah I agree with the rest of the things you said:
-Bless should only roll once
-Humans look for patterns, and tend to notice the bad more than the good.
-Humans are bad at detecting randomness

Fake RNG + Larian Advantage System = Broken Game

I can support this idea. It's already existing code in the game (Social DCs). So for combat it'd be AC - proficiency - modifier. For spell save DC, show the characters spell save DC in the UI.

Interesting results DragonSnooz, my experience differs but I haven't recorded enough to get an accurate distribution.

I was more interested in seeing the difference in low roll streaks since missing a lot in a row is what I was hoping weighted dice would fix and I don't experience much difference in that regard.

I recorded 100 attack rolls against my bear since I can just resummon over and over without other rolls getting in the way, I assume the target doesn't matter.

Here are my results: https://imgur.com/a/9i2rCr3

Non-Weighted Dice
Number of rolls 10 or below: 43
Number of rolls 11 or above: 57
Longest Streak 10 or below: 4 rolls
Longest Streak 11 or above: 7 rolls


Weighted Dice
Number of rolls 10 or below: 47
Number of rolls 11 or above: 53
Longest Streak 10 or below: 5 rolls
Longest Streak 11 or above: 7 rolls


Weighted dice does seem to produce a more balanced roll distribution and an increased number of crits but did nothing to help my unlucky miss streaks.

Thanks Saberem!

It's interesting that we both had high frequency for 17 with non-weighted dice.

I found weighted dice helped the most with damage output. I had fights streak with 1 damage in patch 3, and it's been great to not streak with damage.

This is a good start @Saberem, but you need more than 100 rolls to determine if there is actually a statistical difference between BG3 rolls and randomness. Your non-weighted results produce a Chi^2 of 20.8, and your weighted-results produce a Chi^2 of 24.8. Both of these are below the 95% confidence value of 30.14, and thus we can't conclude that either of these distributions is different than a totally random distribution.

When I add this to @DragonSnooz's weighted-dataset (608 total rolls), the dataset is still indistinguishable from an even distribution (Chi^2 of 16.2, less than the 95% confidence value of 30.14). There are only 2 values that differ from expectation at >3-sigma: 6 (18 rolls out of an expected 30, for a 5-sigma difference) and 17 (40 rolls when we expect 30, for a 3-sigma difference).

...which is a bit odd to me, tbh. This is the weighted-roll dataset, and thus it should be different than an even distribution. Maybe I'm doing the stats wrong?
@DragonSnooz, the "Normal Attack Rolls (Player)" data you posted earlier in this page was for weighted rolls, right?

WHat you have to bear in mind with this sort of thing is that you are part of a population. Going by SteamDB, there are 4200 people playing BG3 right now. That means your 1 in 610 will likely hit 7 of those people tonight, assuming those people only make 7 rolls each and then log out. The probability of them having a run of 7 fails at that level over the course of an average play session is much, much higher.

Oh, does weighted dice affect damage rolls too? The patch notes only mentioned attack rolls, saving throws and dialogue.

If so, this is my unluckiest roll with weighted dice so far: https://imgur.com/a/V85xKXt