so uh, been playing d+d with friends for a while now through dndbeyond+roll20 and something we've noticed is that the dice seem to roll in streaks, meaning that we'll get a bunch of 10s +/-2 in a row, then a bunch of 5s +/-2 , then 18s +/-2, occurring especially with multiple dice rolled at once (ie, advantage/disadvantage). overall, it seems random if you just collect all the numbers rolled together, but looking at the rolls next to each other regularly produces samey~ rolls in sequence. assuming this isn't just confirmation bias and rng ******* with us (which i suppose is possible but it's always been like this for us in the few months we've been doing these weekly sessions), this kind of pattern is something that can be seen in not-as-fleshed-out random number generators. for example, warframe had a similar problem early on, they changed the random number generator such that when rolling a binary choice of black or white (aka 0 or 1), these streaks would not show up and instead produced something that looked more like TV static (in the streaks image, it was very clear something was wrong with the rng with long streaks of black or white regularly emerging).
WITH THAT SAID i also know the random number generator here is more of a physics thing BUT i have noticed some things that would still skew the rolls. firstly and while it's a bit hard to confirm, i suspect that the randomized starting position of the dice is based on one of the faces, rather than treating every die like a sphere and thus allowing it to spawn in any position instead of the presumably 1 in 4/6/8/10/20/whatever positions this kind of spawning would create. this would result in skewing the final rolls due to the "perfect" nature of the physics and positioning involved even with varied throw strength. secondly, there does not seem to be a lot of friction between the die and the "surface," which results in dice sometimes landing on a face and then sliding or sometimes spinning. assuming all surfaces and edges have the same amount of friction, edges and corners should be given a lower+much lower or higher+much higher value respectively so the edges and corners slide or stick more. assuming this is all accurate, which one to go for would depend on the kind of surfaces you are trying to emulate, with lower friction for hard surfaces (where more surface area coverage would be required to increase friction thus having edges/corners "give out" and lose their grip on the surface and slide out from under the center of gravity of the die) or higher friction for sort-of-squishy surfaces like a mat or pad (where the edges/corners can get more easily "stuck" in the material compared to the flat face, thus allowing momentum to potentially have the die change faces). either way would result in more "changes" to the roll of the die and thus allow for it to be more random.
i'm not exactly sure how to translate this into an actual program so as to confirm/rebuff my suspicions myself, but a way to get useful-at-a-glance visual data for this would be to assign colors in order (like a rainbow) to each number of a d20 and print/translate each result as a colored pixel in a paint file sequentially. what you would then have to look for is samey colors grouping together in a row/column (depending on whether you fill rows or columns first). similarly, run this test for several kinds of rolls and die, so 1d20 at a time vs 2d20/4d10s at a time vs 10d20s/100d6 at a time and so on. doing this would allow you to see whether or not such a pattern exists and if it does, which kinds of rolls are the most egregious offenders (ie, the 1d20 set might not have any observable issues but then 10d20 might have a bunch of streaks). if such patterns do not exist, you won't have much, if any, streaks, and the resulting paint files will be cacophonous messes.
using d6s would potentially provide more easily accessible results since you can tie colors to it more easily than a d20 (ie, 1: red, 2: orange, 3: yellow, 4: green, 5: blue, 6: purple) but less potential variation in an individual result might obscure any issues that could exist. all in all, test em all.
so, yeah, uh, is this something that's known of? or are we losing our collective minds over here?
Rollback Post to RevisionRollBack
To post a comment, please login or register a new account.
so uh, been playing d+d with friends for a while now through dndbeyond+roll20 and something we've noticed is that the dice seem to roll in streaks, meaning that we'll get a bunch of 10s +/-2 in a row, then a bunch of 5s +/-2 , then 18s +/-2, occurring especially with multiple dice rolled at once (ie, advantage/disadvantage). overall, it seems random if you just collect all the numbers rolled together, but looking at the rolls next to each other regularly produces samey~ rolls in sequence. assuming this isn't just confirmation bias and rng ******* with us (which i suppose is possible but it's always been like this for us in the few months we've been doing these weekly sessions), this kind of pattern is something that can be seen in not-as-fleshed-out random number generators. for example, warframe had a similar problem early on, they changed the random number generator such that when rolling a binary choice of black or white (aka 0 or 1), these streaks would not show up and instead produced something that looked more like TV static (in the streaks image, it was very clear something was wrong with the rng with long streaks of black or white regularly emerging).
WITH THAT SAID i also know the random number generator here is more of a physics thing BUT i have noticed some things that would still skew the rolls. firstly and while it's a bit hard to confirm, i suspect that the randomized starting position of the dice is based on one of the faces, rather than treating every die like a sphere and thus allowing it to spawn in any position instead of the presumably 1 in 4/6/8/10/20/whatever positions this kind of spawning would create. this would result in skewing the final rolls due to the "perfect" nature of the physics and positioning involved even with varied throw strength. secondly, there does not seem to be a lot of friction between the die and the "surface," which results in dice sometimes landing on a face and then sliding or sometimes spinning. assuming all surfaces and edges have the same amount of friction, edges and corners should be given a lower+much lower or higher+much higher value respectively so the edges and corners slide or stick more. assuming this is all accurate, which one to go for would depend on the kind of surfaces you are trying to emulate, with lower friction for hard surfaces (where more surface area coverage would be required to increase friction thus having edges/corners "give out" and lose their grip on the surface and slide out from under the center of gravity of the die) or higher friction for sort-of-squishy surfaces like a mat or pad (where the edges/corners can get more easily "stuck" in the material compared to the flat face, thus allowing momentum to potentially have the die change faces). either way would result in more "changes" to the roll of the die and thus allow for it to be more random.
i'm not exactly sure how to translate this into an actual program so as to confirm/rebuff my suspicions myself, but a way to get useful-at-a-glance visual data for this would be to assign colors in order (like a rainbow) to each number of a d20 and print/translate each result as a colored pixel in a paint file sequentially. what you would then have to look for is samey colors grouping together in a row/column (depending on whether you fill rows or columns first). similarly, run this test for several kinds of rolls and die, so 1d20 at a time vs 2d20/4d10s at a time vs 10d20s/100d6 at a time and so on. doing this would allow you to see whether or not such a pattern exists and if it does, which kinds of rolls are the most egregious offenders (ie, the 1d20 set might not have any observable issues but then 10d20 might have a bunch of streaks). if such patterns do not exist, you won't have much, if any, streaks, and the resulting paint files will be cacophonous messes.
using d6s would potentially provide more easily accessible results since you can tie colors to it more easily than a d20 (ie, 1: red, 2: orange, 3: yellow, 4: green, 5: blue, 6: purple) but less potential variation in an individual result might obscure any issues that could exist. all in all, test em all.
so, yeah, uh, is this something that's known of? or are we losing our collective minds over here?