[Fluffy]

I for long had suspected that the deal generator that is used for spanish tournaments wasn´t really random.

the reason is that I noticed several times that deal with a low number, was extrangelly similar to a dealwith a high number, ,manytimes with the same pattern, 1 player had some cards at the low level and then was getting a very very similar hand later.

His partner´s cards were trtansfered to an opponent again with similar caracteristics.

This fact wasn´t really noticed by many, of course the reason is that normally nobody cares about high number deals #30-#40, simply because they aren´t played .

I only suspected something wrong was going on, but this weekend in happened something more than 'casual':

Deal #6:

---

And then a for once played deal:

Deal #30:

---

It looks kinda clear that something not random is going on, is it that hard to deal really random hands?

[helene_t]

It is very easy to deal almost-random hands, i.e., hands that are so close to being random that it would take an analysis of millions of hands to discover the nonrandomnes.

It is next to impossible to deal true random hands, but who cares? Even in the hypothetical case that somebody with insider knowledge of the dealing algorithm knew that the chance for a certain finese to work was 49.9% rather than the theoretical 50%, it would have no practical implications.

[Free]

Computers can only deal pseudo-random deals. They always need an algoritm to determine what the next hand will be.

Usually, the more bits you use, the more randomness you'll get. You need some kind of number for ALL hands, then you need to use a formula to calculate a certain number (usually out of the previous number), and that's your next deal.
First of all, to be able to cover ALL hands, you need at least 40 bits. Next you need to calculate with some background number which is a lot bigger, and you use some modulo-method to determine the exact hand.

Pseudo-random generators aren't that easily constructed, since you need a good calculation method before you use the modulo. If that formula isn't suitable, you'll get similar problems, loops,... With such huge numbers, I can understand that it's not easy at all!

[helene_t]

Actually, you would need log2(52!/(13!)^4) = 96 bits to cover all deals, but that's not the issue. The issues are
1) if it is possible to draw any useful inference (such as: the probability that the ace of spades is to the left when I (declarer) have the king is 45%, rather than the theoretical 50%)
2) if it's possible to predict the split given information about previous your own holding, partners holding and maybe even previous deals.
3) if there is a real risk that somebody will encounter the exact same board twice and being able to recognize it.

1) Can be eliminated by analysing the randomness of the deals before releasing the software. It used to be a problem but with todays well-tested randm number generators and excesive computer power for the analysis it is not a problem.
2) This requires, in principle, the 96 bits mentioned for the random seed, and even more if the use of information from previous boards must also be taken into account. For example, if players have access to a history of 100 deals, 9600 bits would be required. That's infeasible. Therefore, in principle, it is possible to cheat if BBO disclosed all details about there software. Fortunately, computers are not available for IRL players.
3) This has been said to hapen with 32-bit software (anybody knows if this is an urban legend?). It should not be a problem with modern software.

[hrothgar]

Few comments here:

First and most significant: Humans are very good at pattern recognition. We're so good at pattern recognition that we often "force" structure on chaos. Its very possible that the patterns that you are seeing are simply an example of projection.

Second: As I understand matters, tournament directors have the option of uploading deals for their tournaments. More information regardig the specific deal generator being used would be very interesting.

With this taken care of, we can start talking about the "fun" stuff:

As Free pointed out, hand generators typically don't using random numbers to generate deals. Rather, they use a pseudo-random number generator as input. The "quality" of the output (the randomness of the deals) can be affected by any one of a variety of different implementation flaws:

A flawed pseudo-random number generator
A flawed hand generator
Poor methods of selecting seed variables for the hands

A comprehensive testing regime would need to subject each layer of this structure to statistical analysis.

There was an interesting discussion a few years ago regarding the application of cryptographic analysis techniques to bridge dealers. At the time, many of the bridge federations were using quite poor hand generator programs which created some interesting "opportunies". In theory, it should be possible to subject a sample of hands to brute force attacks and derive the the original seed variable. In turn, this could then be used to generate the remainder of the sequence. I don't know that any practical examples were ever developed, however, this typoe of analysis is MUCH easier than cracking "serious" algorithms like DES.

For example, lets consider a hypothetical hand generation method for a major torunament like the ACBL Nationals. Assume the following:

1. The ACBL generates all hands for Nationals in advance

2. The ACBL's choice of hand generators is known

3. The ACBL generates all hands for the entire tournament in advance using the same seed

4. Tournament directors apply sets of hands in "sequence". The frist set of hands are used for the first tournament, the second set of hands for thes econd tournament, ...

5. The ACBL uses a perfect seeding technique (for example, shuffle and deal a "real" bridge hand and use this as the seed number for the pseudo-random number generator)

Given this set of conditions, any unscrupulous player could take the set of deals for the first event and use this information to derive all the rest of the deals for the entire event. In this case, the complexity of the attack is limited by the size of the seed variable. The toal number of bridge hands is 52!/(13! x 4). For humans, this is a big number. However, it would be VERY easy for a computer to exhaustively test this space in a very short amount of time.

To me, the logical conclusion is that the set of hands for each session should be generated using a unique seed...

[mikestar]

True random number generation is not possible is software but quite possible in hardware. I don't know all the technical details but I believe than the numbers are derived from various kinds of electronic noise in the computer. A good quailty random number device is around $4000 US last time I saw one advertised. Too much for your local club, but the ACBL could afford one and should use one. (They are sold primarily for use in cryptograpy, but are also used where massive simulations are being done and even an excellent psuedorandom algorithim repeats too soon.)

[rlohman]

We have played more than 100 games on line here.
We have NEVER had a "normal" hand.
Like 1NT-3NT
1S, 2S, 4S

NEVER

Last time the same player got back to back 7 Diamond suits.
Come ON

Random?

Ric

[helene_t]

As for the opening post, I suggest it is a logistic error rather than a fault in the algorithm. Even if they used stone-age software with a 32 bit seed, the risk of two identical boards in the same tournament is small.

[smerriman]

 helene_t, on 2021-January-19, 15:32, said:

As for the opening post, I suggest it is a logistic error rather than a fault in the algorithm. Even if they used stone-age software with a 32 bit seed, the risk of two identical boards in the same tournament is small.

The opening post was 17 years ago, so whatever it was, it's probably not too relevant now

As for rlohman's post, it is random; you like many others just don't understand what random and normal actually mean.

[johnu]

 rlohman, on 2021-January-19, 13:05, said:

We have played more than 100 games on line here.
We have NEVER had a "normal" hand.
Like 1NT-3NT
1S, 2S, 4S

NEVER

Last time the same player got back to back 7 Diamond suits.
Come ON

Random?

Ric

I checked your name against a list of players who are not allowed to get "random" deals. Sure enough, your name is on that list. I don't know what you did but you shouldn't do it again or the consequences will be severe. For a small fee, I may be able to get you off that list.

[blackshoe]

ROFL!