...

[Giordano Bruno 1974]

I daresay that anyone running your games through an analyser is going to find quite a few mistakes.

I already did, though I should probably have explained the results better.

The majority of the players I have faced here are truly awful, some complete novices whose every move is only concerned with keeping every piece from being exposed as a blot. Some are good players. On balance I should be well up because poor play can not consistently stand up to sound strategic play in this game. Luck as a factor should even out and good players must then prevail very significantly. As it is well guess what, I have won a nicely balanced 50% of the games I have played.

Yes, the ELO ranges in which you have played in until now, in large part, only tries to minimize blot exposure. Which will usually result in exceptionally poor error ratings.

Yet, your own error rating isn't great either. I plotted your last 10 games into GNUBG which resulted in these overall ratings.

Error rate mEMG (MWC) __________ -37.2 ( -0.152%)
Snowie error rate _______________ -15.1 ( +0.000%)
Error based abs. FIBS rating _______________ 1666.7

GNUBG file

Error rate (per decision): The error rate per decision is the total error rate divided by the number of non-trivial decisions (i.e., the sum of unforced moves and close or actual cube decisions). Again, please note that is different from Snowie 4 that defines the error rate per decision as the total error rate divided by the total number of moves for both players. In general, your error rate per move will be lower in Snowie than in GNU Backgammon. An investigation of approximately 300 matches showed the on average the GNU Backgammon error rate with be 1.4 times higher than your Snowie 4 error rate.
Equivalent Snowie error rate: For easy comparison GNU Backgammon will also print the total error rate calculated by the same formula as Snowie 4.

Threshold for ratings:

Normalized total error rate per move Rating
0.000 - 0.002 ___________________ Supernatural
0.002 - 0.005 ___________________ World Class
0.005 - 0.008 ___________________ Expert
0.008 - 0.012 ___________________ Advanced
0.012 - 0.018 ___________________ Intermediate
0.018 - 0.026 ___________________ Casual Player
0.026 - 0.035 ___________________ Beginner
> 0.035 ________________________ Awful!

Please note: The Normalized total error rate per move is multiplied by 1000 in the default settings.

The games that were plotted in were played by CaractacusPots' on 2020-09-03 and 2020-09-04.

CaractacusPots' overall error rating within the 10 games that were plotted in is within the 'Awful!' rating.
The lowest error rate within the 10 games was -13.7 and the highest -71.7.

In relation to BGA Backgammon ranking it is probably around 200 ELO.

Yes, I read that at the time. Though you were polite enough to not describe it as "awful".

I meant the three games he played and lost recently against people with higher ELO level than him.

In his mind, he was pwning the opposition who then suddenly got lucky because of the "in-built levelling". According to his arguments about weaker players getting help, he should have been the one getting help.

My view is that if anyone were to run an error count, I'm sure that at least one of them had a lower error count than he did.

[Giordano Bruno 1974]

I'm not going to do this

b) because I don't want to, I have no doubts that the RNG function works perfectly fine

You can't possibly know that the RNG function has been implemented "fine" unless you've run or seen a proper analysis of a huge sample size of dice rolls. Your statement is as bad as mine suggesting that the RNG may be off. Without the appropriate analysis of a meaningful sample size no-one knows.

c) because it isn't going to change anything - you've already been given plenty of reasons, statistical data etc but you ignored it all so you'll do the same no matter what

Not true. If I were to crunch the numbers of a large sample and it all looked ok, right number f doubles, right number of individual dice throws etc etc then I would be pacified. There seems to be a reluctance to provide the data though.

I think it's fine because I play several games that are quite chance based e.g. yahtzee and I haven't felt that the results are odd. So I have no reason to start examing data. Other people have already done this and were apparently satisfied. Added to which this site is very open and transparent, I can look at the source code if I want.

People have already suggested ways that you can investigate the existing data that you already have. The number of doubles from your games is about right. Your issue is that you seem to think that 16.6% of all throws should be doubles each game. This is a fallacy. I suggested to you that you needed to calculated the expected standard deviation in order to compare it against the actual standard deviation. I don't think you know how to do this. Plus you'd only play the "well they gave me false data" argument anyway.

Also I don't remember who said it but it has been pointed out to you that your preselection of dice rolls argument would only work if the game knew in advance how many dice rolls there were going to be (which is clearly impossible) so your existing data set should be sufficient.

[Jest Phulin]

BGA already has a method of any player to replay any game, something I doubt the online poker site had.

You'd be wrong. All major poker sites maintain full logs of all your games automatically on your own hard drive so you can analyse the data...

Um, being able to analyze one's own hand isn't the issue that I was doubting. It was the ability for anyone to replay a game by any player. You don't have to have been involved in the game in any way to watch the replay.

One piece of information that you seem to be missing, too, is that the replies so far have not been from BGA staff. You have made several mentions that "the information is in in your database." Nobody that has responded has those databases. Some of the people who have responded are volunteer developers, so they can more easily see the code. But anyone can sign up to be a developer and see the code as well.


A curiosity, too. Why do you want a large sample size, when your first statements were that the site was using random numbers that would pass randomness checks, but assigning them in a manner to throw the game to the weaker player?

[euklid314]

And another game is stolen away



Once again I have opponent pawned within a handful of moves. I have a near perfect wall, my home quad has 5 primes blocked off and I've knocked off opponent to the bar. I will likely fill the open 2 point in the next 2-3 moves and then waltz home. But the system has other ideas !

Opponent has just a 1 in 3 chance to roll a 2 to get back on.

His first roll is 3 3

His next roll is 2 3, and back on the board he goes
I knock him off again and he gets straight back on with 2 6
I knock him off again he gets gets straight back on with 2 3
I knock him off again he gets 4 5 so his one man stays on the bar
I knock off another blot he has left open so now he has 2 men on the bar
He rolls 2 1 and gets one man straight back on
He rolls 2 1 again and gets second man on the bar

This is just nonsense sorry.

He should throw a 2 to get back on 1 time in every 3

Out of 7 attempts to do so he succeeds 5 times.

Consequently I go from a totally winning position to losing.

CaractatusPots, thank you for posting 2 positions where you are convinced that the outcome was "strange".

I try to lead the discussion to something more constructive. Note, that some week ago I have analyzed a single game from you, CaractatusPots. I have mentioned it in an earlier post of this thread but my contribution was lost in the discussion. But beside analyzing this single game I have read several of your posts which reveal something about your playing attitude, your strengths and your weaknesses.

In my opinion you probably play the opening well, the middle game ok/average and the end game is your weak point. If you will focus your training on the end game you can improve your game a lot!

I see two major weaknesses (which you can overcome!):

1) If you are in a good position you overestimate your winning chances and feel too safe. If you are in a bad position you underestimate your winning chances and resign, even if you still have a realistic chance to win the game.

Please try to estimate your winning chances in your example that I have quoted above: You have a 5-point board, your opponent is on the bar and lets assume it is even your turn to move next. We both agree that you are on a winning path. Your winning chances are obviously somewhere between 50% and 99,9%. Try to guess your winning chances (please assume perfect play of yourself and your opponent). Please do not use computer help - because in a live game you also do not have computer help. I assume you will overestimate your winning chances.
Then try to guess your chances if you have the same situation in the screenshot but assume that it is your opponent to move next. Will your chances to win now be significantly different - or can we neglect this aspect?

2) You make checker errors especially in the end game. In other words, you are unfamiliar with some of the techniques to bring a favorable game home as safely as possible.

Please read books especially on the doubling cube (because you learn a lot about win probabilities in certain situations) and read books about end games. Your game will improve significantly.

P.S.: I hope many readers of this thread are interested in the example too and you can all try to estimate the winning chances of the red player. Lets separate your guesses in A) red to move and B) yellow to move. I will give the solution to this puzzle after several people have given their guesses.

[CaractacusPots]

2) You make checker errors especially in the end game. In other words, you are unfamiliar with some of the techniques to bring a favorable game home as safely as possible.

Not quite Euklid. It's simply that once I know I've been rogered by unfair dice rolls I have no appetite to try and win the back game at that point. By that point I have simply resigned myself to the fact that I have had the game stolen from me, a game I would comfortably win the majority of times from that position. I can accept that opp will get his blot onto the bar 1 time in 3, but when he's knocked off 7 times in that position and he gets on 5 of those times I tend to simply give up.

Sure, yes, I could fight, I might somehow turn it around but by that point I just don't have the appetite for it. I know that with REAL DICE I am beating this guy the vast majority of times from this position. But game after game after game when I'm ahead the outside anomalies magically happen to re-balance the game and give the underdog a fighting chance from what would otherwise be a rather hopeless position. There's just no fun in playing when this nonsense happens with monotonous regularity. It just doesn't happen with such frequency in real life just as river river miracles in online poker don't happen anywhere nearly so often in real bricks and mortar casinos.

The problem here is that every denier can only offer up the rather lame excuse that "well, anomalies happen otherwise it wouldn't be random".
Problem is it happens all the time. I mean 18 doubles in one single game? In 50+ years that has never ever happened to me in real life games.

There are lots of possibilities here. Lots of ways the anomalies could be being generated.

The only way to be sure of what is going on is to analyse a significant subset of the BG histories here.

And unfortunately I cam't get hold of that data.

If I could I would gladly number crunch it.

[euklid314]

Please give me a number on my question 1)

How hopeless do you see the situation if you were player yellow in your screenshot? If it were your turn would you fight with real dice? How large would you estimate your winning chance with color yellow?

The answer to this question is crucial to my argument. Please note that I "know" the "correct" winning probability. As you probably know one can let the computer play millions of times from this position (the computer plays super-humanly strong with each move, so the dice will decide the outcome). How many games out of 100 will yellow win?

[CaractacusPots]

Please give me a number on my question 1)

How hopeless do you see the situation if player yellow in your screenshot? If it were your turn would you fight with real dice? How large would you estimate your winning chance?

The answer to this question is crucial to my argument. Please note that I "know" the "correct" winning probability. As you probably know one can let the computer play millions of times from this position (the computer plays super-humanly strong with each move, so the dice will decide the outcome). How many games out of 100 will yellow win?

I would expect to lose in opponents position probably 7 times out of 10 but the manner in which those losses occur matters.

In that position if opponent just gets ONE lucky roll of 2-2 he's on the board, makes a point in his home quad which makes it difficult for me to get back men out and so on. That wouldn't be so bad TBH.

But whatever YOU think the win rate is from that position (and I believe it is significant) the problem remains that after he gets on the board and I knock him off the 2nd time, my win rate from THAT position is also very significant. And then when I knock him off a third time it is again very significant then again from the 4th, 5th and 6th knock offs to the point where I knock 2 of his men off in that position and at that point I am probably winning over 90% of games from that position. It's just a nonsense.

I would hazard a guess that any attempts to tweak outcomes of backgammon games is potentially very complex and difficult EXCEPT for 2 specific instances which are surely very easy to code. Once instance is when you're knocked off the board and the system therefore knows exactly which dice throws will keep you on the bar or let you off. The other is simply the number of doubles a player is given as more doubles equals more PIP score and more chances to make important points. Interestingly it is precisely these two situations where my games appear to be being constantly impacted.

[euklid314]

Thanks for you answer.

You estimate that red wins 70% and yellow wins 30%. Your guess was a very good one.

The true numbers are that red wins 80.8% of the games and yellow will win 19.2% of the games. So in every 5th game you will lose the game with red EVEN WITH PERFECT PLAY.

Please note that in practically all of these 19.2% of the games something crazy must(!) happen. The chances for yellow to win are very low (I am surprised that you estimate it at 30%) so he needs some very, very lucky rolls. In approximately 20% of the games you will wonder with awe what true random numbers can produce! True random numbers generate crazy patterns if you love searching for patterns!

Just to give an example: The probability that from the screenshot position yellow wins even a backgammon is 3.16%. So if you start a million of games from the position, then every 33th game yellow will win the game and red will not even bear a single checker off! I am quite sure that you would be shocked if you are in this commanding position with red and lose a backgammon. But even that will happen quite frequently even if you play perfectly. If your moves are suboptimal and your opponent plays better then you might lose 5-10% of these red games even with backgammon and 30-40% of the games in total. Note that it is way easier to play on the defensive side - every professional will make more errors if he/she is in an offensive position compared to an defensive position.

Summary: If you use random dice then in approx. 20% of the games yellow will be incredibly lucky and win the game. In approx. 20% of the games red will roll incredibly lucky rolls but you will never realize that because you expect red to win and most other rolls will win anyway. In 62% of all games red will even win the game with a backgammon - but you will not be surprised.

Again I come to my conclusion: You expect a "true random sequence" to go back and forth in a nicely fashon - with only minor surprises every now and then.
Often I hear that such crazy things do not occur while playing with real dice. But that is completely wrong and shows either misjudgement (our mind is very strong at focussing at things we want to see) or lacking experience. One only has to view ANY longer backgammon match on the internet (say a match to 17 points) and one will ALWAYS find SEVERAL such crazy outcomes and turnarounds in single games of the match. These backgammon tournaments are played with precision dice - which produce the best of random numbers one can get with physical dice.

[CaractacusPots]

Thanks for you answer.

You estimate that red wins 70% and yellow wins 30%. Your guess was a very good one.

The true numbers are that red wins 80.8% of the games and yellow will win 19.2% of the games. So in every 5th game you will lose the game with red EVEN WITH PERFECT PLAY.

Please note that in practically all of these 19.2% of the games something crazy must(!) happen.

I think your definition of "crazy" differs from mine. I don't personally think anything "crazy" needs to happen for opponent to win the game. A simple 2-2 (1 in 36) roll is enough to really change his chances. But once he fails to get on, then I get to block the remaining open point and it's mostly game over at that point. In the event his being dealt a 2 to get on the board 5 times in 7 when his chances should be only 1 in 3 is the ridiculous or "crazy" part of the game. What that is in essence is a 3 sided die (if such a thing could exist) rolled 7 times and a 1 coming up 5 times of those 7. Many people would believe that there must be something hooky with the die if that happened.

Here's the position a little later in the game:



I'm way way way ahead again in this game. Pip count is my 95 vs his 151 and I have 2 of his men knocked off in that same 5 point board.

Perhaps you'd care to calculate how many times I win from this position.

Yet system deals him a 2 on his next 2 consecutive rolls.

I only need a 2 myself to cover that remaining point or a 6-4, 55

It's just so off in some way but unless I can get to the history logs of all the games I can't prove anything.

[Jest Phulin]

It's just so off in some way but unless I can get to the history logs of all the games I can't prove anything.

A method for this has already been presented. Yes it is tedious. Almost as tedious as responding to your accusations.

I only need a 2 myself to cover that remaining point or a 6-4, 55

There are 36 ways to roll 2 dice. 11 of them contain at least one 2. 1 of them is 55. 2 of them are 6-4. Giving 13 of 36 ways for you to get what you want. And 23 ways to not get it. Roughly speaking, you are twice as likely to not get the roll you wanted than you were to get "I only needed ..."