After 20 times playing, I am done with the BGA version

[KillbotFactory]

But I'm all for using a "better" source of randomness for BGA, just because it will prove my point that these threads will continue regardless.

Yep. Three things in life are certain: death, taxes, and RNG complaints.

At least the third one is entertaining!

[Stroom]

It could still be coincidence (couple ten thousands rolls are not enough, to say anything non trivial). But claiming it being pure coincidence generally, without anything backing you up (on the objective side, in math/statistic sphere [not the psychological stuff; about, why people see stuff/patterns that isn't there, and why people complain about the "non randomness" - because with this part youre right, afaik]), is just as wrong to claim its 100% not coincidence - you just need data (either from actual games that bga COULD provide, or by running the algorithm that the devs used for catan), and some adequate tests for it. In the time we discussed it, one could easily write the code for it, if that access/information is given.

"Catan has too few rolls to provide a fair distribution." - never said that the distribution looks off .... (in the end of post games it just looks unsuspicious , like most looked everywhere, more or less)
"The game IS quite random so you have to deal with it." - without actually testing that, we just waste words; my aim was to help out the dev team to make the game better (still in alpha tester mode)... I should probably let this be, until either the devs react (bcs. bga wont^^) or it will stand out in competitive play in a couple of seasons, and finish my UI-Rework suggestions

You can't prove that "something isn't there". Therefore, YOU should prove that there actually is an issue.

The way you want 10k rolls to show something... but then say that the distribution itself seems OK... shows that you don't really know about how RNGs work or how to test these things. Or even how probabilities and statistics work.

If you want to prove it, you have to show that there are some specific sequences and that you can predict when and how they happen. And that you can predict each roll before it is rolled. If you can't do that then you are wrong. I can not prove that there isn't a problem. How do you prove that? Even if I do any kind of test, there would be someone else who would say "but that was not how you should test this, try X, Y or Z method too". And then it just keeps on going. So... YOU should prove based on your information that there actually is a real problem. Predictable outcomes, not just that some rolls keep happening in clusters. 100% predictions please.

[Romain672]

Please try to give something specific. So something like:
"if two succesful players roll two 3s on a row, then the probability to roll 3s in the whole game would be higher than normal"
That way, we can test that hypothesis, and see if it's true or not.

Atm, I think you are just saying that one number will happen too much, but there is 11 differents numbers which can lead to that.
You could said
"if two similar number are rolled by two succesful players, then the probability to roll that number in the rest of the game will be higher than normal".
But then, if this is true, that mean there will be a big difference between what you expect in average on that exact roll.

I'm not talking about a number of roll on each number like 3333533 (since on that '5', you got the start of that hypothesis, when you started lookind for that number) but something like 3333733 (so something like 3333533 after starting looking for that number) which will be a big difference.
If I'm not clear on that last sentence, just ignore it. Rest of my post should still be clear.

[silentProtest]

It could still be coincidence (couple ten thousands rolls are not enough, to say anything non trivial). But claiming it being pure coincidence generally, without anything backing you up (on the objective side, in math/statistic sphere [not the psychological stuff; about, why people see stuff/patterns that isn't there, and why people complain about the "non randomness" - because with this part youre right, afaik]), is just as wrong to claim its 100% not coincidence - you just need data (either from actual games that bga COULD provide, or by running the algorithm that the devs used for catan), and some adequate tests for it. In the time we discussed it, one could easily write the code for it, if that access/information is given.

"Catan has too few rolls to provide a fair distribution." - never said that the distribution looks off .... (in the end of post games it just looks unsuspicious , like most looked everywhere, more or less)
"The game IS quite random so you have to deal with it." - without actually testing that, we just waste words; my aim was to help out the dev team to make the game better (still in alpha tester mode)... I should probably let this be, until either the devs react (bcs. bga wont^^) or it will stand out in competitive play in a couple of seasons, and finish my UI-Rework suggestions

You can't prove that "something isn't there". Therefore, YOU should prove that there actually is an issue.

The way you want 10k rolls to show something... but then say that the distribution itself seems OK... shows that you don't really know about how RNGs work or how to test these things. Or even how probabilities and statistics work.

dude, for freakin fucks sake, dont you even read the links posted ?!
"You can't prove that "something isn't there"." - ofc, you can proof it is random enough!:
{Frequency Test: Monobit
Frequency Test: Block
Runs Test
Test for the Longest Runs of Ones in a Block
Binary Matrix Rank Test
Discrete Fourier Transform (Spectral Test)
Non-Overlapping Template Matching Test
Overlapping Template Matching Test
Maurer's Universal Statistical Test
Linear Complexity Test
Serial Test
Approximate Entropy Test
Cumulative Sums Test
Random Excursions Test
Random Excursions Variant Test}
from : https://www.random.org/analysis/ - guy on the first page posted it - no distribution does not NEED be a factor if you just can guess the next result, its not random enough ! (as in sequences "423156423156.." might be one, but it it really doesnt need to be a trivial one - and all pseadorandom results are some kind of sequence. and there are methods to test if that's unpredictable enough..)

... on your second part, I dont quoted - you suggest I find out how to cheat here, without the tools that could make it fun .... and share that ... for free, to be fixed ? - ofc I will

Edit:
PS: a way to see youre not really interested to look futher in, is that youre actually don't read relevant links - I very openly said, in the bug report, that I will try to change my meta, to see if that fact is worth exploiting; and if it is I will very openly do that ^^
PPS: I see where my mistake was - I should have just opened a new thread - plugging my own bug in the middle and expecting people to read it was probably stupid^^

Edit for clarification: I'm referring always to this, and not something vague and non-concrete (and not to the original post of the thread)! : https://boardgamearena.com/bug?id=66082

Last edit, because I wont bother to make more posts here, if it's not interesting or actually worth it - I probably answered to "that" critique in some previous post, in this thread, just (re)read that one ( because same stuff/complaints/statements start to be looping)

[Stroom]

You still don't seem to know what you are talking about. This is an analysis of thousands of rolls. If you scale it down to 70 rolls per game, these things just do not matter. If you can't predict these sequences and prove it, then there is no point in trying to implement any kind of more expensive RNGs. It's also probably much slower and takes up more resources in the server. Rather just be OK with what is faster and cheaper and STILL not predictable.

Catan players are like this in every implementation. Do you think that they all use such a broken RNG? No. It's all confirmation bias.

[Romain672]

Here is 40 randoms generations of 60 rolls: https://www.zupimages.net/up/22/29/xu8k.jpg
A1 look normal (no 2s though).
B1 is way off on 12, you get a big cluster there.
C1 got a cluster of 5s
D1 got a cluster of 5s and 10s
E1 got a cluster of 7s and 9s
F1 is way off on 6s
A2 is a little normal, but slightly off in 4s and 7s
B2 is way off in 2s
C2 is off in 9s
D2 is off in 9s
E2 is off in 7s
F2 is normal.

What I want to show with that: you find clusters everywhere. On my 12 randoms games, 10 got some clusters. But everything is normal: we don't have enough rolls to have a good distribution.

You think your distribution is weirder than those SilentProtest?

[silentProtest]

Here is 40 randoms generations of 60 rolls: https://www.zupimages.net/up/22/29/xu8k.jpg
A1 look normal (no 2s though).
B1 is way off on 12, you get a big cluster there.
C1 got a cluster of 5s
D1 got a cluster of 5s and 10s
E1 got a cluster of 7s and 9s
F1 is way off on 6s
A2 is a little normal, but slightly off in 4s and 7s
B2 is way off in 2s
C2 is off in 9s
D2 is off in 9s
E2 is off in 7s
F2 is normal.

What I want to show with that: you find clusters everywhere. On my 12 randoms games, 10 got some clusters. But everything is normal: we don't have enough rolls to have a good distribution.

You think your distribution is weirder than those SilentProtest?

OHH , I see the confusion now !!

what I mean by clusters is on a time-result-diagram - omg, hope that wasnt the communication problem all along (but i think i explained it better in the report) .... what i mean by cluster is some thing like that (lets use [1-6] with 1/6 each): 123561235644441235612356 ; this would be a 4 cluster - and if this happens once in a game this would happen prob. again later with same number ( not rarely with same combination of red and yellow dice number) - THAT is what I find curious, and way to often that it would be just random (obv. Id need a larger data set too proof that - super unlikely sequences can just happen - but it would be nice to have a number to put on, how unlikely that consistency is... )

PS: had games where numbers where 3-4 times as often as it was likely - but its not super weird, with that few rolls, and I thought nothing about that

[Stroom]

Again confirmation bias. If you take all the digits of PI and search for "12345" then you will find very many clusters like this.

Also... Show us your games where this happened.

Also... this is not how RNGs work. If the RNG seed was like this, then it just happens.

[Jellby]

Here is 40 randoms generations of 60 rolls: https://www.zupimages.net/up/22/29/xu8k.jpg

C3 is obviously not random, your RNG is broken.

[Romain672]

Ok so now, we can reproduce it.
Per example, is here what I propose you to do from now for us to start believing you:

Each time you roll twice the same number, you count the next roll of the game.
Then you see if you rolled that same number one more time or not.
(you can decide before starting the stats to count 6/7/8 only if you roll them thrice or not)

Once you get 30 rolls (if you started at any point of your historic, like from now), you can give us results on what you find. It's possible that you would need a total a 100 rolls to start seeing something, but 30 rolls should give us a good idea.

edit: And give results into that form:
3s: 0/2 sucess
4s: 1/3 sucess
5s: 4/4 sucess
and so on.