After over 500 games I am done with the BGA version

[Tharnduil]

I mean I feel the same way and am often playing to just see what absurdities arise each game.

But with that said I do recognize that I am prone to remember the crazy instances that work against me and less so the ones that work in my favor. In truth as others have said Catan is just a broken game. You either just play the randomness straight up, or if it's live or more conducive to chat then you are trying to manipulate bad players. Both are not the way for strategy games to be quality.

With all that said, I do feel like the random number generator is broken on this site and should be revisited. Here is my last game:
https://boardgamearena.com/table?table=337195448

37 total rolls
2s - 0 times
3s - 4 times
4s - 4 times
5s - 7 times
6s - 1 time
7s - 2 times
8s - 3 times
9s - 7 times
10s - 7 times
11s - 2 times
12s - 0 times

What makes it more absurd is that the 5s, 9s, 10s one player had them for all but one roll. Again the issue is that this doesn't seem like an exception as opposed to the norm. I feel it both when it works for me and when it doesn't.

Too many times it feels like the number generator plans out the percentage averages they are supposed to reach and then you get crazy runs of certain numbers early that even if they end up evening out by the end of the game it is usually too late if the player is some what competent.

So I think both things can be true. Catan is a broke game vastly subject to the RNG of the dice, and this site may want to relook at the system they use to randomly generate the numbers.

[Stroom]

Another case of human bias. Only 37 rolls. Randomness works like this.

[SwHawk]

Specifically : clustering illusion and possibly Texas sharshooter fallacy

[Tharnduil]

Way for both of you to completely miss the point as I noted that fact in a "one off" as well.

Like I said if you read the post instead of just jumping to conclusions, the issue seems to be that these seem to be the rule instead of the exception. I mentioned both that we are drawn to pay attention more to ones that work against us rather than for us. But the fact remains that these "irregularities" seem to happen quite regularly on this site.

I am not claiming, broken! or Unfair against me! as others have done, but I do think that the question at this point is a legit one when I have seen multiple posts (some reputable some less so) with data for irregularities. To just pass it off as stupid humans being stupid is also a 'pretty silly take as that is how things don't get better.

To me it seems, like I posted, that the RNG that is used here tries to hit a percentage as opposed to being truly random. Too often in games it seems that early one number, or set of numbers, will hit giving a huge edge (I believe another poster mentioned this in the past). Then towards the end of the game the numbers will start to hit their percentage level.

And if a player is good then you can exploit that early numbers to an impossible advantage. For another example in my favor here is my last one that I won without rolling more than 1 seven and only getting one Dev card vp point:
https://boardgamearena.com/table?table=337501660

I finish the game in just 31 rolls!
the numbers 4, 5, 6, 9 all exceed there expected totals and all happen to be my most important numbers.

Again the argument isn't that these things are impossible. The argument is that these type of games that should be irregular seem to be common at best and expected at worst.

[Romain672]

Here is my last game:
https://boardgamearena.com/table?table=337195448

37 total rolls
2s - 0 times
3s - 4 times
4s - 4 times
5s - 7 times
6s - 1 time
7s - 2 times
8s - 3 times
9s - 7 times
10s - 7 times
11s - 2 times
12s - 0 times

I didn't wanted to do it initially but since you insist.
I took the % of chance to roll exactly those number of time those numbers for each roll possible (per example there is 35,26% chance to roll exactly no 2), then I added all of them, which gave me 1,589.
And I did 3000 simulation of 2x37rolls with libreoffice.

I find that 90 games out of those 3000 (3%) was below that number.

Remember that the thing you checked was done AFTER you see your game, and it's a pretty weird one. You are not looking for number of doubles or one specific number being rolled too much or two few, but you was looking for something weirder.


But I don't like my post, so I did something else.
I search for the third most lowest number between those 11. It give 6,286% which was your number of 9s.
Out of 3000 others random generation: 22 were below that number so would be even weirder. 11 were equal (so their third number which 'was rolled the most weider' was 9 or 5 and a total of 7 times).
That give 1.1%.
But again, I was looking for something you claimed after your game.

If instead I do something fun to only check the number which was first in 'rolled the most weirder'. Here it would be 10 with 2,11%.
With 3000 simulations, I got 696 which was more unlikely that your game, and 89 equal.
That give 26.2%.

So if I'm choosing the thing we are looking for, I find 26,2. It look pretty much a random number between 0 and 100, isn't it?


Please be precise in your claim. List all the unlikely event you think of. Then come back in 100 games, and show us all the game which got one of those events And then we could do something with it.

[Stroom]

With 2 6-sided dice you will have 36 combinations. Similar to winning in 37 rolls.

If you look at 1 set of coin flips, it needs 2 throws to get every combination. So the possible combinations are (0,0), (0,1), (1,0), (1,1)... So if you look at all these results, most players would say that (0,0) and (1,1) are unfair. 50% of the game rolls seem unfair.

It is a VERY simplified way to look at it but I'd argue that with two 6-sided dice, the issue is very similar - Out of all possible outcomes that can happen with 2 dice, 50% of the games you would not be happy with (if the game ends at around 36 rolls). The chance might decrease in 70-80 rolls (normal winning time in most balanced games) but not much. Maybe to 25%-33% but don't count on it.

You still talk in a very conspiracy-theorist way which indicates human bias.

Just play the game and do not take it seriously. I'd rather remove elo from this game as it really does not show much about the real skill of the players. Fluctuates too much.

Also... You seem to only play 3-player games. These are very simple and it is easy for one player to get ahead early and win with some early lucky rolls. Catan is much more balanced with 4 players.

[Tharnduil]

First off, thank you both for a more reasonable response than the others. Want to say that because it is appreciated.

Secondly, again I'm not claiming it's broken or that it's biased against me, in fact it worked in my favor just that last game I listed. What I am saying is that this claim has been posted multiple times, both by some people with credible claims and others that are just frustrated with an RNG game working against them.

I'm saying that I have noticed some irregularities not over a "one off" but it seems to be more the rule. Those last two games I posted are back to back. In my last 5 I think one of them had a more even distribution of numbers and the other 4 all seemed highly irregular. Again I don't care. I get Catan is an RNG game, especially so on the 3 player map which frankly is just a broken game full stop.

What I am saying and what you suggested Romain is that the game devs should do a deep dive of their RNG based metrics in a 100 game set, print the results to the community, and waylay some of these consistent claims that happen all too often (or fix the system if it proves broken). So even from your numbers, Romain, it is clear that while not impossible they do seem a bit irregular, but within an expected happen stance of a value set of games. And that is my point. If there are multiple claims here on this site, isn't it the responsibility of the Devs to make sure their system is working properly?

So as you articulated, "Please be precise in your claim." My claim is thus:
- this issue has been touched on too many times both with some credibility and with random humans frustrated by RNG and not understanding how it works.
- I have noticed irregularities that seem to be the rule instead of the exception.
- I do not know if this is actually just a random sample size of the games I've played in opposition to the numerous thousands of ones on the site, but numbers do seem to cluster in irregular patterns and others have noted it so that raises a (if not red at least yellow) flag
- It would be appreciated for the devs to run their system on the game over a large sample size, put the finds in a quick write-up so that when, inevitably this issue comes up again any poster can just link that studies findings and be done with the conversation as the argument has then been settled.

Again what I am not claiming:
- that there is for sure an issue
- that I am angry at an RNG based game for being well RNG heavy
- that there is any rubber-banding as the first poster touched on as that seems pretty absurd

Finally, I would much prefer 4 player Catan, but for some reason the "Gurus" don't think anyone wants to play 4 player Catan so the Arena is closed off to 4 player Catan. However, I do join the tournaments from time to time and just got 2nd place and I believe if I stuck to only 4 player Catan my ELO would probably reflect 400-500+ with my game results.
Sadly, I enjoy the selection of games through Arena and instead just have a good chuckle in any 3 player Catan match as I go into it expecting the most broken of games as it is extremely RNG heavy.

[euklid314]

I also would like to stress the fact that several players think that the expected distribution (1/36 of the rolls should be 2s, 2/36 of the rolls should be 3s,...) should be noticable in a game of Catan. This is not the case!

Since a game of Catan is so short (40-60 rolls) it is highly unlikely to approximate the "perfect" distribution within a single game. As Stroom already mentioned, you have 36 combinations of 2 6-sided dice. The chance that these 36 combinations show up on the first 36 rolls is (practically) zero. Of course in Catan rolls are clustered (since 1-4 is the same as 4-1 and is the same as 3-2, etc.) but still 40-60 rolls is not nearly enough to expect anything near the "ideal distribution". In fact the dice would be rigged if the ideal distribution would come up exactly in any single BGA game of Catan.

If the dice are random, there is a guarantee (!) that the numbers will be clustered in some way. And these clusters are to expected to be really noticable and "surprising". If after one game the feeling is that the dice were fair this time than this will be a huge exception! The size and amount of occuring clusters will of course be different in each game and only on the long term (say 100-1000 games) one can judge if the dice rolls are correct.

Also, what Tharnduil phrased with "Too often in games it seems that early one number, or set of numbers, will hit giving a huge edge" shows a common misconception. Players often assume the first rolls should show numbers different from each other. Why should they? If the rolls are independent from each other, why should they care of what was rolled before? Again it is hugely likely to have clusters at the beginning of the game which even out a little bit during the game. This is just the so-called "law of large numbers".

If you roll your dice the first time in a game of Catan, and you roll a 9, then the probabilities at this moment are completely broken, since you have 100% 9s in your game so far and 0% on all other numbers. Everybody expects that phenomen and has gotten used to such a huge outlier (since it happens every game so nobody cares. But if after 5 rolls 3 of them are 9s and 2 of them are 3s, then most people feel uncomfortable because the distribution seems so weird. The distribution of 60% 9s and 40% 3s in the game is still much (!) better than it was after the first roll and it will get better and better (i.e. closer to the expected distribution) during the course of a game and will even get be getting better and better if you watch 1000 consecutive games. Again "law of large numbers".

One further remark on those players that claim that these outliers do not happen in their real life games or not in that extent: Those players usually do not write down all their rolls during their real life games and therefore do not collect any data of their dice throws. And thus they have no tools to analyse their data for their last game of their last 100 games. At least I do not personally know anybody who collects such data in friendly RL games...

[euklid314]

In answer to your most recent post, Tharnduil, which was written in parallel to my most recent post:

Please read what the FAQ of BGA have to say on random numbers:

https://boardgamearena.com/faq?anchor=f ... ing_random

Short summary and personal remarks: At the moment BGA uses the PHP-function random_int() which is used any many applications world-wide. If there were problems with this RNG (= random number generator) than we would hear from it on every news channel (newspapers, TV,...) since more important areas would be influenced by those problems than only a board game site.

BGA is also implemented in a way that the current state-of-the-art RNG can easily be replaced by a better algorithm if it should become available.

Existing RNGs (like random_int) produce random numbers that are at least as good as those that can be produced by physical precision dice. If you don´t trust the mathematical models that prove that fact then you might try to prove the contrary.

BGA did all it needed to do by using the PHP-function random_int() and documenting this to us users. They only need to ensure every game developer uses this function - which they do.

[Romain672]

It's one of those problem to do no claim about what you call 'weird'. That let you come up with anything you want (like rolling no 2s during 16 games in a row (which have a probability around 0.1%), and then claiming that part.
If I want to do 'proper' stats with it, I would then need to write down everything you could claim during your game (like maybe 200 differents claims if you count all differents numbers), and so at this point, if I said that X have 1% chance of occuring, since 200 events can occur, I can transform that probability into 86.6% (1-(1-0,01)^200).
And so when you see that 1% you can said to yourself it's weird since it happen once every 100 games, but in reality, you just get one of the 200 possible event, which make one of those 1% event happen in 86.6% of cases.

I will repeat, but the fact you don't list those make it very hard to do anything with future claims. I'm not asking you to do it (esp if you don't plan to take notes about 100 games), but to convince anyone who know stat with 'one random game', you would need events 'not too weird' and being maybe in the 1/1000M range. Or two events in the 1/1M range (0,0001%).