Dice is not random at all

[Mart]

I have some serious doubt about the dice. I just played a few games in which it was nothing but 5's and 9's. Hardly any 7 or 6 or 8 was rolled. When I look at statistics after the game, a normal distribution cannot be seen in the times a certain number is rolled.
This game is far from perfect. It laggs, it crashes occasionally, but I at least expect that i can play the game in a fair way.

[N o i r]

Publicado originalmente por sparky:

This were true if I asked the question: "How likely is it that we get the exact same distribution again". However, the question I'm asking is: "when rolling the dice 30 times, how likely is it that any single 2-12 result is repeated 8 times". In true random, and an infinite sequence, any sequence of 30 rolls will adhere to this probability. This is a mathematical fact, do you agree?

Yes. In any random environment, the possibility of repeating X times in Y rolls any number Z between 2-12 can be calculated beforehand.
I'm not sure what you mean by "in an infinite sequence".

Publicado originalmente por sparky:

I've said as much to others and I completely understand your point: sequences of numbers are irrelevant when dealing with any single random unspecified event.

I'm assuming you mean that the result of previous rolls do not affect the outcome of the new ones; if so, yes, this is also correct.

Publicado originalmente por sparky:

Again, you are right in general, but not with PRNGs, because with PRNGs you're not dealing with a single random unspecified event, but with all generated numbers by the PRNG. PRNGs are judged by 3 things:
1. likelihood that each number appears as many times as probabilities suggest
2. no patterns, correlations or predictability between the results
3. the time at which each number appears in the sequence is consistent with probabilities with high confidence.
You're continuously thinking of 1. and 2., but you're never considering 3. Do you understand 3. ? Because again, this is mathematical fact. And I'm not implying predictability, but rather if you roll the dice N times and 7 doesn't come up, the likelihood of 7 not coming up in the next 5-10 rolls goes up, in a completely random scenario. That's why roulettes at the casino have betting limits (which causes many to lose their life savings)

We do not know if Catan Universe uses a PRNG, a HRNG or what else. But even with a PNRG, I doubt that sequences ("streaks") of rolls would be significantly more prevalent than with other types of RNG. As in, I have not seen evidence showing that the possibility of repeating X times in Y rolls any number Z between 2-12 is significantly different between PRNG, HRNG and true random, for any range of those parameters.

Publicado originalmente por sparky:

Exactly, you shouldn't play this game with real dice. But let me tell you that if you were using Catan's PRNG, I would definitely play this game with you, me as the player, and you the casino. Because this type of repetition happened too frequently for me.

Let me try to explain my point more clearly. I just went to random.org and generated 20 rolls:

7 3 5 7 3 7 8 6 7 7
8 6 2 6 8 8 5 10 3 4

After examining this sequence, I concluded that:

-the possibility of never generating "12", "11" and "9" in 20 rolls is 0.6%;
-the possibility of having at least 5 times "7" and 4 times "8" is... I didn't do the math, but I assume it's similarly rare;
-the possibility of having 5 times 7 in the first 10 rolls is 1.3%;
-the possibility of having the first three factors at the same time is even less likely;
-or for what matters, the possibility of achieving those exact 20 rolls in the exact same sequence I just rolled them is astronomically low. And yet it happened.

In your case, you choose to examine the chances of a given roll in X times, but you did this after the fact; it is very easy to find an unusual property in a sequence after it is already rolled. Almost every sequence will have unusual characteristics.

If you try to measure the likelyhood of those specific events, you will most likely find out that indeed, they occur as rarely as they should: the chance for 2 to come up 6 times in 10 rolls 1 in 100 million, 3 coming up 8 times in 25 rolls is 5 in a million, etc.

In other words, those rare occurrences will happen as rarely as they are supposed to be, and just because they happened to you that time doesn't make the RNG a bad one. It does not imply that the distribution is flawed.

Publicado originalmente por sparky:

And where I'm getting at with this is a consistent method for judging whether a PRNG is good or bad, based on this simple 8-time repetition test: Take a PRNG and get N results (N>1,000,000). If the number of sequences of 30 dice rolls showing any single number (2-12) repeated at least 8 times (if 2 numbers are repeated at least 8 times, then the sequence is counted twice) is greater than 0.15 * N, then the PRNG is bad.

Do you understand and agree with this test? Because then we can generalize this to include even rarer situations like 3 repeating 10 times in 25 rolls, and see how many sequences with such low likelihood, exist over N rolls. Can we agree that failing these tests will definitely prove (at least for the purpose of our conversation) that the PRNG is bad? And if these sequences don't appear as much as I am suggesting, then, obviously, my argument would have no basis.

Yes, it would a good test. You calculate the odds of a given event ("repeating a number 8 times in 30 rolls"), then you repeat the 30 rolls N times, and for N sufficiently high the outcomes should get expontentially close to the prediction. This by considering the series of 30 rolls as a single instance of the event.

Publicado originalmente por sparky:

Fair enough, except you also have to justify why it's invalid, and so far it's not that clear to me. I've given extremely unlikely scenarios, happening in close proximity, as well as fairly unlikely scenarios, happening much too often for my games.

1. Is it because you assume it just so randomly happened to me a few times, but wouldn't happen again or to others as often, thus admitting that these numbers make the PRNG look at least suspicious if everyone had similar experiences? The assumption of low likelihood you're making is contested by many players' experience and testimonies, and also by the fact that it keeps happening to me

2. is it because you don't believe the scenarios? I've provided screenshots, and I can play 2,3 catan games, and I can almost guarantee at least 4 or 5 1-in-100 scenarios (at 50-60 rolls per game, in 2-3 games should only get 2-3 one-in-100 scenarios on average), and at least a 1-in-10,000 scenario (which is absurd if I can actually predict this), so we can do this experiment if you'd like

3. is it because you don't believe sequences matter in assessing a PRNG's success? Again, I can go over this and explain why sequences and the frequency of various sequences matters. This again is a mathematical fact which I can easily prove.

4. is it because you disagree on the mathematics of why these low likelihood scenarios should only occur, on average, once in many games? Again, I can elaborate on this and present more math to back it up.

My answer is that you haven't recorded nearly enough of your games to measure whether or not the occurrences of those specific events is "too common".

1. is close to my answer. And no, this does not make the RNG look suspicious, for the reasons explained above.
3. also applies. A sequence is just a combination of rolls whose likelyhood of outcome can be measured with precision; it follows the same rules of any other statistical event, and is subject to the same kind of tests.

A good test would be calculating the % of, say, the number "3" repeating 10 times in 25 rolls (which is 37 in 1.000.000); run this test 1.000.000 times, and notice that the number of times this comes out is not too far from 37.
Then for good measure, run this test 100.000.000 times, and notice that the number of times this comes out is not too far from 3.700.
Because this event is particularly rare, the number of tests required to get close to the expected outcome is quite high.

Publicado originalmente por sparky:

The proof is easy, let's do the 8 repetition in 30 dice test (which is not the lowest likelihood scenario I've seen, the 3 repeating 10 times in the first 25 rolls is a much stronger proof, but let's use the frequent events instead):
- if we assume a 60 dice rolls per game, and play 10 games => 600 dice rolls
- we record the first 30 dice rolls in every game
- We then count the # of times any number is repeated 8 times.
- this gives us the answer to a simple probabilities problem: given 10 sequences of 30 random dice rolls each, what's the likelihood any of these will have any number repeat 8 or more times? 15% as you saw in the casino test.

Math, statistics and probabilities say, in true random, we expect this to happen 15% of the time (any of the numbers repeated 8 times over the first 30 dice rolls of a game).
- We have 10 of these exclusive uncorrelated sequences => approximately 1.5 numbers will get repeated 8 times in the first 30 rolls, over 10 games. This means in 10 games, you should see a maximum of 2 numbers in total (not per game) repeated 8 times in the first 30 dice rolls. By pigeonhole principle, in at least 8 games out of 10 on average, you shouldn't see this kind of repetition in the first 30 dice rolls IF the PRNG is good.

=> Since I've had 2-3 numbers repeating 8 times in the first 30 rolls every 3-5 games or so (sometimes more often, and sometimes multiple numbers repeating like that in the first 30 rolls of a single game - I have screenshots), it means this expectation based on probabilities does not hold with Catan Universe, hence the PRNG is bad ... And you'd probably take my money in the casino game above, if I was the casino running the dice on the Catan Universe algorithm.

Looking forward to your reply

You want to play 10 games (where each game is at least 30 rolls long) and verify that the chances of any number happening 8 times in 30 rolls converges to the expected result?
We could do that, but keep in mind that just 10 repetitions of a statistical event is not a number high enough to grant that the results will be close to the expected frequency.

[noob master]

Publicado originalmente por author="N o i r":

After examining this sequence, I concluded that:
...
In other words, those rare occurrences will happen as rarely as they are supposed to be, and just because they happened to you that time doesn't make the RNG a bad one. It does not imply that the distribution is flawed.

Yes, I know this, and I addressed this already... I'm not asking the question: "How likely is it that I get the same repetitions in this sequence". I'm saying "How likely is for N sequences that ANY number repeats 8 times in 30 rolls". The first is a silly question that you and I both know is unrelated to probabilities, the second is something that should hold for any random algorithm.

For example, turning your question about 9 not appearing over 20 rolls in a generic one applicable to PRNGS, I would ask: in 1000 games, how many have 9 not showing in the first 20 rolls. Then I go ahead and play 1000 games and check. If I played N games in which specifically 9 didn't show in 20 rolls (i.e. the same number is missing) then we compare with the math expectation of this. This is a good test for whether the PRNG favours or disfavours a specific number.

Publicado originalmente por author="N o i r":

Yes, it would a good test
...
You want to play 10 games (where each game is at least 30 rolls long) and verify that the chances of any number happening 8 times in 30 rolls converges to the expected result?
We could do that, but keep in mind that just 10 repetitions of a statistical event is not a number high enough to grant that the results will be close to the expected frequency.

Yes, it's very likely the pattern I've seen here won't be repeated. But that's not what I've seen in my games. I've kept trying to come back, and numbers repeating has been a problem consistently in the fist 30 rolls. I think it's worth doing as a test on Catan Universe. If not by me, by the Universe developers.

In addition, there are plenty more generic questions we can ask about the distribution, to assess it's adherence to true randomness & probabilities:
- How likely is it that in the first 40 rolls of a game, ANY single number doesn't appear? How about the probabilities for any number between between 4 and 10 not appearing? This second question should again also happen once in 100 games or less. However in less than 10 games I've seen 5 not show in multiple games, and this does keep happening a bit too often (especially as now we have a decent definition of what "too often" means).
- How likely is it that in the first 30 rolls of a game, ANY single number appears more than 10 times (instead of 8)? How often should I see this?

I'm asking and measuring results for these questions separately. And again the odds from math don't match the small sample size I've dealt with.

Again, I understand this may well be anecdotal. However I also understand it may not be.

[N o i r]

The likelyhood of any given number to repeat 8 times (or at least 8 times) in 30 rolls is a statistical event, distinct from that of other events.

So if you have "p2", "p3", ... "p12" (where pX is defined as "the chance of rolling "X" 8 (or 8 or more) times in 30 rolls, each of which can be reliably calculated using the equation you kindly provided a couple pages back), then:

the chances of NOT rolling any one of these sequences is:

(1-p2) * (1-p3) * ... * (1-p12)

the chance of rolling _at least_ one of these sequences is:

1 - ((1-p2) * (1-p3) * ... * (1-p12)).

[fearenough]

the chance of having fun playing this game 0

[MSISR]

Yes, this game is build purposefully without random dice rolling.

[N o i r]

Publicado originalmente por MSISR:

Yes, this game is build purposefully without random dice rolling.

No, it's not.

[noob master]

Publicado originalmente por N o i r:

Publicado originalmente por MSISR:

Yes, this game is build purposefully without random dice rolling.

No, it's not.

Yea, I doubt it's on purpose. More like a willful bad design decision not worth the investment for them. And dice for catan are what cars are to racing games. You don't invest in the cars (design, control, etc.), you'll get a crappy racing game. Same with Catan which is why many have quit the game, and some of those are posting in here. For the record I uninstalled Universe months ago and have only played colonist.io since.

Also, @N o i r, look at this link https://www.random.org/analysis/ . You may believe me or not, but what I said before was actually from my math training not from this site, but they're saying the same things I was, that any mathematician knows: it's impossible to prove a given sequence of any numbers was or wasn't generated randomly. What instead we do is count the statistically unlikely events, and match them against probabilities. If you get a 0.003% situation happening 2% of the times, it's clearly a bad PRNG.

For example, any single one of the 11 roll numbers missing in the first 35 rolls happening more times than expected => bad PRNG. And this should be done with all "unlikely" statistical events, such as:
- how likely is it that any number is missing in N rolls
- how likely is it that all 11 rolls appear in N rolls
- in the first N rolls, how likely is it to roll 5 consecutive numbers, (i.e. rolling 2,3,4,5,6 or 5,6,7,8,9)
- and so on.

Oh and there's another thing I forgot to mention in the past, which again is why catan universe is horrible: all of these statistics have to hold over EVERY PLAYER'S GAMES. You can't say: "it evens out in the long term for all games on the server", if a few players get too many 9s, while others get too few 9s, for example.

Realistically, a true Random Number generator will inherently adhere to ALL of these probabilities. EACH AND EVERY ONE. With minor statistical deviations.

And in the "Visual Analysis" part of https://www.random.org/analysis/ you can see the visual difference between true randomness, and very poor randomness, the likes of which I saw in the Catan Universe dice. It's ridiculous really. I know I've said this before, but myself and all others in this thread who had issues with Catan Universe, have felt the dice patterns in the game be unrealistic, way too often. Some like you stick your head in the sand and say "we just gotta trust the random gods", but when you do the math it's just not there.

[N o i r]

Publicado originalmente por sparky:

Same with Catan

What do you mean with this?

Publicado originalmente por sparky:

Also, @N o i r, look at this link https://www.random.org/analysis/ . You may believe me or not, but what I said before was actually from my math training not from this site, but they're saying the same things I was, that any mathematician knows: it's impossible to prove a given sequence of any numbers was or wasn't generated randomly. What instead we do is count the statistically unlikely events, and match them against probabilities.

I'm aware of this, but there is one thing that can be done: measure a statistically signicant (that is, large) amount of samples and see whether or not it asyntothically converges to the expected median.

Publicado originalmente por sparky:

If you get a 0.003% situation happening 2% of the times, it's clearly a bad PRNG.

Only if the numebr of tests is sufficiently high. If you ran 50 tests and in one of them you happen to stumble upon the 0.0003% chance, this does not mean it's a bad PRNG.
From the very link you quoted, "This means that a good random number generator will also produce sequences that look nonrandom to the human eye".

Publicado originalmente por sparky:

For example, any single one of the 11 roll numbers missing in the first 35 rolls happening more times than expected => bad PRNG. And this should be done with all "unlikely" statistical events, such as:
- how likely is it that any number is missing in N rolls
- how likely is it that all 11 rolls appear in N rolls
- in the first N rolls, how likely is it to roll 5 consecutive numbers, (i.e. rolling 2,3,4,5,6 or 5,6,7,8,9)
- and so on.

All the examples you brought have a statistical chance of happening that can be calculated. The chances of a "2" not happening in 35 rolls is exactly (35/36)^35, which is 37%. This is almost 1/3. "happening more times than expected" yes, this is the key. The sample must be large enough for suspicion of bad PRNG to be substantiated.

Publicado originalmente por sparky:

all of these statistics have to hold over EVERY PLAYER'S GAMES. You can't say: "it evens out in the long term for all games on the server", if a few players get too many 9s, while others get too few 9s, for example.

Realistically, a true Random Number generator will inherently adhere to ALL of these probabilities. EACH AND EVERY ONE. With minor statistical deviations.

I'm afraid I don't understand what you mean here.

Publicado originalmente por sparky:

I know I've said this before, but myself and all others in this thread who had issues with Catan Universe, have felt the dice patterns in the game be unrealistic, way too often. Some like you stick your head in the sand and say "we just gotta trust the random gods", but when you do the math it's just not there.

What makes you say that? "Feeling" that a pattern is unrealistic is no statistically sound evidence, you should know that. And of course, my position is not the one you highlighted.

[turton.michael]

Yes, the RNG is broken. It falls into predictable sequences that are easy to spot. Last night another with my girlfriend, it was 4-6 most of the game, especially early. Almost no 9s were rolled. This is a common pattern, where two numbers are far more frequent and one number, usually more common, disappears almost completely. Usually as soon as the locked pattern becomes discernible I try to get on those numbers for the easy win, but my gf was already there and I could not, and I lost predictably. After that I quit playing, the RNG is so awful that I can seldom tolerate more than a game.

We always laugh when I call out "6!" or "4" and it is. So easy to predict.

[N o i r]

Publicado originalmente por turton.michael:

Yes, the RNG is broken. It falls into predictable sequences that are easy to spot. Last night another with my girlfriend, it was 4-6 most of the game, especially early. Almost no 9s were rolled. This is a common pattern, where two numbers are far more frequent and one number, usually more common, disappears almost completely. Usually as soon as the locked pattern becomes discernible I try to get on those numbers for the easy win, but my gf was already there and I could not, and I lost predictably. After that I quit playing, the RNG is so awful that I can seldom tolerate more than a game.

We always laugh when I call out "6!" or "4" and it is. So easy to predict.

Still waiting for a detailed writeup of this magical prediction algorithm of yours.

[fearenough]

Publicado originalmente por N o i r:

Publicado originalmente por turton.michael:

Yes, the RNG is broken. It falls into predictable sequences that are easy to spot. Last night another with my girlfriend, it was 4-6 most of the game, especially early. Almost no 9s were rolled. This is a common pattern, where two numbers are far more frequent and one number, usually more common, disappears almost completely. Usually as soon as the locked pattern becomes discernible I try to get on those numbers for the easy win, but my gf was already there and I could not, and I lost predictably. After that I quit playing, the RNG is so awful that I can seldom tolerate more than a game.

We always laugh when I call out "6!" or "4" and it is. So easy to predict.

Still waiting for a detailed writeup of this magical prediction algorithm of yours.

your ego make you into ..... you just dont see whats stated. IT IS IMPOSSIBLE TO PREDICT THE 2 NUMBERS TILL THE GAME HAS ROLLED 20 ODD ROLLS AND THE SAME DAMM NUMBERS ROLL OVER N OVER AGAIN BUT Y FIGHT A EGO FILLED CLEVER GUY WHEN I DONT PLAY THIS SUCK GAME NO MORE

[fearenough]

here goes what a dev with nothing to hide would reply like the risk devs:Originally posted by Zoundz:
Is there and advantage to having the special dice? Im noticing Im getting some consistent bad beats...I would like to buy the upgraded dice...where can I find more info?

I can confirm to you that the dice skins have no impact on gameplay and are purely for aesthetic/bragging rights.

We have uploaded our dice code onto github if you'd like to do a thorough inspection

https://github.com/smgstudio/risk-dice

[fearenough]

Publicado originalmente por fearenough:

here goes what a dev with nothing to hide would reply like the risk devs:Originally posted by Zoundz:
Is there and advantage to having the special dice? Im noticing Im getting some consistent bad beats...I would like to buy the upgraded dice...where can I find more info?

I can confirm to you that the dice skins have no impact on gameplay and are purely for aesthetic/bragging rights.

We have uploaded our dice code onto github if you'd like to do a thorough inspection

https://github.com/smgstudio/risk-dice [/quotehave to say risk is a game with some really skilled players that kick my ass alot but i can learn as i have but with a game with bad dice i can not learn all i can do is get pissed off n if you look at my play time almost stop playing all together

[N o i r]

Publicado originalmente por fearenough:

Publicado originalmente por N o i r:

Still waiting for a detailed writeup of this magical prediction algorithm of yours.

your ego make you into ..... you just dont see whats stated. IT IS IMPOSSIBLE TO PREDICT THE 2 NUMBERS TILL THE GAME HAS ROLLED 20 ODD ROLLS AND THE SAME DAMM NUMBERS ROLL OVER N OVER AGAIN BUT Y FIGHT A EGO FILLED CLEVER GUY WHEN I DONT PLAY THIS SUCK GAME NO MORE

In other words, your theory is untestable, unverifiable, unfalsifiable.

[N o i r]

Publicado originalmente por fearenough:

here goes what a dev with nothing to hide would reply like the risk devs:Originally posted by Zoundz:
Is there and advantage to having the special dice? Im noticing Im getting some consistent bad beats...I would like to buy the upgraded dice...where can I find more info?

I can confirm to you that the dice skins have no impact on gameplay and are purely for aesthetic/bragging rights.

We have uploaded our dice code onto github if you'd like to do a thorough inspection

https://github.com/smgstudio/risk-dice

The developers of Catan Universe have repeatedly stated that their RNG algorithm is fair:

https://forum.catanuniverse.com/topic/7996/the-dice-are-not-random/2
(scroll down for the dev post)