Poker Players Alliance Forums » Poker Players Alliance

Is the DealGuardian engine biased?

(17 posts)
  • Started 3 weeks ago by Big Jim Slade v2.0
  • Latest reply from nmaiorana

  1. Big Jim Slade v2.0
    Moderator

    I feel bad about posting this here... however it pertains to a real study about potential fraud in online poker rooms. I wanted to spin this off into a new thread.

    Someone please tell me this is not the place for it. Please :)
    BTW - minor spelling corrections were made.

    About a week ago, eBlade posted:

    I ran across this little PDF "analysis" from Nick, that he posted to a blog I frequent.. seems he's going all around the Internet looking for anyone discussing cheating at poker, and posting things to advertise for himself.

    http://www.securecarddealer.com/presentations/BigHandAnalysisReport.pdf

    He's basically accusing all of the sites out there of cheating, pending his analysis of online hands. Unfortunately [edited], I doubt it's pretty likely that his analysis of online hands will ever come out to be large enough, because he can't witness the hole cards of every person every hand (I hope :D ), so therefore, he'll only see pocket-pair-races in the situations where people just shove it to showdown.

    Here's a totally non-scientific study for you, though -- I'm a dealer at a card room. In a 7 hour shift, I'll usually see AA vs KK 2 to 3 times in a night. I deal about 20 hands an hour, more on tight or short tables. I've dealt AA vs KK vs QQ three times (that I'm [edited] aware of) in the almost year I've been dealing. I've witnessed AA vs KK vs QQ with K and Q both hitting the flop, twice. (queens won the first time hitting quads on the river, and aces won the second time, hitting a set on the turn) (both of those times, I was a player, so I didn't include those in my dealing)

    A specific player will be dealt AA one in every 220 hands. A specific player will be dealt KK one in every 220 hands. Give or take a few, SOMEONE will see AA or KK (or any other specific pocket pair) on a table, if you were able to see everyone's hole cards, once every 22 hands, if full handed.

    My conclusion: Nick's [edited] simulation is probably not real, and his numbers don't seem to add up to real-world mathematics. If I'm incorrect there, please tell me, because I'm certainly not a math wiz, although I am a computer programmer, so if Nick [edited] would post his simulation code, we can check it for flaws.

    The above was from eBlade, not me.

    Posted 3 weeks ago #

  2. Big Jim Slade v2.0
    Moderator

    In the above referenced DealGuardian simulation, 56 million Texas Hold'em starting hands were created and tallied. In this study it is reported that AA occurred 4.07% of the time, and TT occurred 4.08% of the time. There is also a statement saying that the expected mathematical value is supposed to be 4.08% for all pocket pairs, not just pocket tens. This is what the plain language of the pdf clearly states.

    While eBlade claims he is not a mathematician, non mathematicians may not realize that the 0.01% variance on 56 million poker starting hands is not an expected difference of two or three hands, but rather 5,600 hands.

    In other words, Nick from the DealGuardian company is stating that his DealGuardian card dealing engine will short pocket Aces about 5,600 times over the course of this simulation. In other words, you get pocket Aces from DealGuardian less than you would from other dealing engines employed by online card rooms, and less than what you would expect to see in a casino.

    I don't know what Nick's intentions were, but this is what he said in the paper he is circulating.

    For those of you who are not Mathematicians, let me say that when 5,600 occurrences of pocket Aces are missing, you might want to go looking for them. In a truly random deal (and by that I mean a computer generated "natural randomness") you will see the exact same number of pocket Tens as you will pocket Aces when considering a large corpus of hands. There are 4*3 = 12 possible pocket Aces out of 52*51 = 2,652 possible starting hands. For the uninformed, 56 million is a sufficiently large corpus.

    While we still have no real information on this from Nick at the DealGuardian company, in another thread I suggested that DealGuardian possibly uses a Fisher–Yates shuffle. Nick will neither confirm nor deny that his product uses the industry standard shuffling algorithm acknowledged by all to be the best. I have pointed out the two most common problems in the implementation of the Fisher–Yates shuffle - index and modulus. I've also previously pointed out that these type of implementation errors would produce fractional percentage errors for some cards, such as Aces.

    All I can say is that by my reading of the BigHandAnalysisReport from Nick at the DealGuardian company the DealGuardian product clearly produces a biased deal by failing to deal 5,600 pocket Aces in the simulation.

    What do others think?

    Posted 3 weeks ago #

  3. kovera
    Member

    It would seem to me that a .01 variation is completely acceptable when discussing a 'random' shuffle. After almost 10 years of playing poker online, I have never felt like the cards are favoring one player over another. Random chance is always a big part of poker. In fact, if it were only a question of mathematics, anyone would be able to do it and the game wouldn't be worth playing

    Posted 3 weeks ago #

  4. TheEngineer
    State Director & Forum Admin

    Someone please tell me this is not the place for it. Please :)

    It's fine here, Big Jim.

    Responses are fine and expected, so long as company reps don't promote their products.

    Posted 3 weeks ago #

  5. Big Jim Slade v2.0
    Moderator

    Kovera makes a point. Just because the shuffle from DealGuardian is a biased deal, there may be no reason to not use it. It's someone's personal preference. I suppose if I was at a home game and one of the players just couldn't shuffle and deal competently, I would not bother to excuse myself and go home.

    I took a second look at the 0.01% number. That is the variance on pocket Aces, not on the deal. Since (according to the folks at DealGuardian) the expected occurrence is 4.08 percent, the actual deal bias is 0.01/4.08 = .00245 = 0.245%, or about 24 times the number we first looked at.

    So the deal is biased for about a quarter of a percent for Aces. Your potential for getting an Ace is less than for getting a Ten when DealGuardian provides the cards.

    Now admittedly, in any small sample of a Monte Carlo simulation you will get some variance. In examining 1,000 sample hands you expect variance. But this was not a small sample. The sample of DealGuardian hands was the equivalent of playing online poker for eight hours per day Monday through Friday (holidays excluded) for 392 years. Technically, because we would say in a true simulation of this sort "orientation is not important," this is really a simulation of 3,529 years of daily poker playing.

    Mathematically this error (bias) is known as a "limit" problem. As the number of sample hands approaches infinity, the deal bias in a Monte Carlo simulation will approach zero. While we never quite reached an infinite number of hands, in physics or engineering 3,529 years of daily poker play is technically referred to as "close enough for all practical purposes."

    In other words, this is not variance in a small sample - the deck is simply stacked.

    The Brick and Mortar equivalent of this is going into a casino and being told there is only one seat at one poker table left. You can have the seat but we want to tell you that the dealer likes to stack the deck by pushing Aces to the bottom to remove them from play, and he puts Tens on the top of the deck. Of course he doesn't do this all the time, and might not do it today. But it is the only poker seat left and we can't change dealers anytime soon.

    The only conclusion I can come to using Deal Guardian's own data is that with DealGuardian the deck is stacked.

    Posted 3 weeks ago #

  6. nmaiorana
    Member

    Just to be clear, DealGuardian is more than just a virtual card shuffler. It's real purpose is to provide extra security. Security that would have prevented UB and Absolute.

    I used the DealGuardian engine to shuffle hands in my study because it's readily available to me. But the reason I did the study is to show how often AA-vs-KK should come up during a poker session. I see it probably twice an hour playing online and this seemed a bit excessive for me.

    Instead of talking probabilities and statistics (which could have mathematically determined how often this situation occurs) I decided to simulate a 9 player table's starting hands. Each sample simulated 8 hours of play at a rate of 70 hands/hour. In the end the big pairs (AA, KK, QQ, JJ, TT) came up with the statistical average they were supposed to. This demonstrated the accuracy of the DealGuardian shuffle engine.

    Then I looked at how often these big pairs met each other as starting hands. I calculated a rate of once every 9 hours, this is in line with the statistical average. There were sessions that saw them more frequently, but on average it was once every 9 hours.

    So why do I see them so often online? Unfortunately, there is no good way for me to get credible online data and analyze it. Just keep your eyes open for what you are experiencing online. Any feedback is greatly appreciated.

    Posted 2 weeks ago #

  7. thunderacura
    Member

    The reason you see the big pairs at a higher rate is that way more hands are dealt. You can't just go on the 70/hour. If you consider all the concurrently running tables, it's actually way more than 70 hands per hour. It's quite possible for you to see AA multiple times within an hour where someone at another table doesn't see AA at all their entire session. I play online everyday and have experienced both sides. I can even recall a live game in a casino where I saw AA 5 times with in a 2 hour time frame and 2 of those times the person right next to me had AA as well. I think for argument to be valid you would have to see AA and "excessive" amount of times every time you play online

    Posted 2 weeks ago #

  8. nmaiorana
    Member

    AA did come up the statistically accurate number of times (along with KK, QQ, JJ and TT). The study gave us a rate of around 3 per hour for each. All 5 of the measured pairs approached the statistical probability of 4.08% It's when another player has KK, QQ or KK -vs- QQ, the study came up with once every 9 hours or around 0.15%. Any of the three big starting hands should be seen around once every 3 hours.

    I used the 70/hands per hour based off of what I was seeing at 9 handed tables. What rate do you think would be a better value? I can provide the raw data from the study if anyone is interested.

    Posted 2 weeks ago #

  9. Big Jim Slade v2.0
    Moderator

    In all fairness, nMaiorana, in your published study for the Secure Card Dealer DealGuardian, you never actually achieved the established statistical probability for Big Pairs.

    In your study you used the DealGuardian Engine for a simulated three millennia (3,000+ years) of playing poker and failed to converge on the expected statistic for AA, KK, and QQ. You were close, but you did not converge on the proper statistic.

    It's a limit problem. As the number of hands approaches infinity your Monte Carlo simulation should approach an error of zero. The error will become nearly unmeasurable in a short period. After a simulated three and a half millennia of play, you still had an error of 0.245% in your deal for all large pairs.

    To put this in perspective - if the rake was short by this same percentage on this same number of pots, it would be missing quite a few dollars. 56 million pots (orientation is not important) times $4/pot * 0.245% = $548,8000.00 in lost rake.

    Speaking on the "study," nMaiorana said:

    ...there is no good way for me to get credible online data and analyze it. Just keep your eyes open for what you are experiencing online. Any feedback is greatly appreciated.

    You did ask for feedback. Your dealing engine is mathematically biased. True it is a small bias, but it shows you can't implement an unbiased deal, so the rest of your software is suspect. This same attention to detail shows up in your analysis and promotion.

    In your "study," you talk about comparing as is to expected. You include spelling errors in the published PDF suggesting your work, in general, is sloppy and you don't double check your work.

    From your published PDF on Big Hand Analysis

    ...and compare theme to the expected results.

    (emphasis added) Apparently you meant to say "and compare them to"

    These spellings errors, your inaccurate statements about poker rooms, and general sloppiness are pervasive in all your writings where you tell us to trust you because you don't make mistakes in your product.

    When asked to provide a warm and fuzzy feeling to the consumer by assuring us you use industry standard best practices along with well known established Off The Shelf technologies for these standardized problems your reply is that you use secret technologies - trade secrets - rather than the established dependable solutions that others have employed.

    Just like for an unbiased deal, encryption technologies are simply an expression of a string of random numbers. If you have problems with deal bias that you cannot resolve, it stands to reason that your encryption is flawed as well.

    You say there is no way for you to get credible online data and analyze it. I don't see why not. Others do it. The AP/UB scandals were outed by analysis of online data - these were poker players. The recent famous 76% no showdown number was based on analysis of, as I recall, 103 million online hands - this was a poker quality assurance company (isn't this similar to what your company is?). What quality do you lack that prevents you from accessing the same data that others had, or gathering data from similar sources?

    You are allegedly a company that is selling a "better" set of cards. Don't you have contacts in the industry that can supply you with the data you say you cannot get? I suspect some weakness of mind that has you claiming you cannot get real online data. Is that weakness of mind in your head, or is it a supposed weakness of mind in the heads of the players because you think they will believe you when you say readily available data is not available to you.

    You shun third party verification of your product. You claim you should be allowed to exist in a "wild west" of online poker where no one has oversight or jurisdiction over your company and it's practices.

    You ask us to put the hidden cards of the poker room in your hands, because only your hands are honest. You claim you want to prevent the occurrence of another AP/UB scandal all the while saying that while you currently cannot get to the data to do this, if we put the same data in your hands you won't peek.

    If I recall correctly, your DealGuardian server benchmarked at about 400 hands per hour. Using your own statistic of 70 hands/hour that is less than six tables per server. If any poker room wanted to use your servers, then it would take a huge number of your servers to implement your product.

    I regularly toss women into the air, high enough to hear them squeal before I catch them in my safe arms. But despite any alleged strength I have, I would not trust your product as far as I could throw it.

    Posted 2 weeks ago #

  10. nmaiorana
    Member

    The statistical error you mentioned above was just a bit high. It was less than 0.0074%, clearly approaching 0. The data demonstrated no bias.

    Don't forget that if it was not for a lucky break, UB and Absolute would have continued. Players received information that is not part of a standard operating procedure from customer support from those sites. That's not likely to happen again.

    And once again you are missing the point. Our product would have prevented those situations, and yet you still are denying the need for such a product. Without it it's players -vs- the poker sites when it comes to scandals. The poker sites will win most of the time, unless their support staff breaks their rules. People just like you will tell players that they need to improve their skills and there losses are their own fault.

    That's what happened prior to the last scandal being exposed. Yet, it was not bad play, it was players getting cheated. Bottom line, there is a problem with allowing sites to maintain active hand information during play. DealGuardian solves that problem and is good for the player and the site. Why is this not a good thing?

    Posted 2 weeks ago #

  11. Big Jim Slade v2.0
    Moderator

    The discussion with Nick about his DealGuardian product has developed in ways I did not expect. So in conclusion I have to step back and ask myself what is the lesson here?

    Some years back when Planet Poker's shuffling algorithm was shown to be severely flawed, they made changes and fixed the problem. In contrast, as I have shown Nick that his DealGuardian shuffling algorithm is flawed, he has merely quipped back that all is well and there is no problem here.

    The problem is not so much that the software is flawed, but rather that the company won't acknowledge their errors. Nick has over 24 years experience in information technology. He has worked in software development and systems architecture. Nick is in charge of product development at Secure Card Dealer for their DealGuardian product. It is incumbent upon Nick to understand the algorithms and math behind the software he is creating and selling.

    Deal bias is an error in the deal expressed as a mathematical "limit" on an infinite series. ( A concept from High School Calculus.) Based on data provided by Nick at Secure Card Dealer the DealGuardian bias is a "limit that approaches 0.245%". The limit of the infinite series does not approach zero, it approaches 0.245%.

    Nick says

    The statistical error you mentioned above was just a bit high. It was less than 0.0074%, clearly approaching 0. The data demonstrated no bias.

    Ok then, you have no concept of the math if you say the limit approaches zero. You provided the data, not me. If the error is not 0.245%, but is instead 0.0074% as you now say, then it is still not a limit approaching zero. It is a limit of an infinite series approaching 0.0074%.

    So where does your new number, 0.0074% come from? If this a restatement of the 4.07% frequency of pocket Aces? Does this mean the frequency was not really 4.08% - 0.01% = 4.07% as stated in your paper, but really 4.08% - 0.0074% = 4.0726% ? If so, then then deal bias is not 0.0074% but rather 0.0074% / 4.08% = 0.18% deal bias. So your bias is not a mathematical limit approaching zero, but approaching 0.18%.

    Nick - run a chi-square test on your data. Your two-tailed P-values are off the chart! There is simply no way your published results came from a fair and unbiased deal. With DealGuardian, the deck is stacked.

    Nick, how do I say this without making it a personal attack? Your deal has mathematical bias. You don't know what you are talking about, or you do understand the math and are lying about it.

    For the reader who is not well versed in the math, you submitted empirical numbers to simulate over 3,000 years of playing cards every day. You deal bias converged as a limit within about 30 days of virtual dealing – by your original numbers converged on 0.245% bias for pocket Tens over pocket Aces. Then for the next 3,000+ years of virtual dealing stayed at the same limit it originally approached. You will, with your current algorithm, never approach a limit of zero on the infinite series of DealGuardian deals, even if you deal for another 3,000 years.

    Why not do what Planet Poker did? I gave you information on why your deal was flawed, why not fix it like other poker rooms have done in the past?

    Respectable poker rooms will display their third party oversights. The poker rooms are regulated to some degree or another by various authorities. It might be a license with the Isle of Man Gambling Supervision Commission. They might by a member of the Interactive Gaming Council. The poker room might submit to be audited by companies like Citigal or BMM International.

    Respectable card rooms will describe the entropy sources used by their Random Number Generator (RNG). They will name the cryptographic hash algorithms they employ and the bit lengths. They will describe their deal and shuffle process explaining the use of the RNG and encryption.

    DealGuardian does not do any of this. There is no third party oversight. There is no DealGuardian specifications of how known established and trusted technologies are employed. There is no description of the general inner working of the shuffle like a respectable card room provides.

    What DealGuardian wants, and wants badly, is to get their paws on your hidden cards and have them in their offices in North Carolina. Nick wants, as an original insider of DealGuardian, to be able to have control over your down cards and the upcoming board cards. Just like in the Ultimate Bet scandal that racked up over $22 million for the company insider there.

    I'm not saying Nick will look at the cards - after all Nick says only he is honest among all worldwide card room employees and he will not be looking at your hidden cards when he plays poker on the same card room. But the capability is there. Maybe his receptionist's boyfriend will look at the hidden cards. Who knows?

    As laws are developed to provide consumer protections in the area of online poker, should just the poker room company be regulated, or should third party vendors be regulated as well to prevent unscrupulous companies such as Secure Card Dealer from introducing flawed products such as DealGuardian?

    Posted 1 week ago #

  12. nmaiorana
    Member

    I will not be accused of trying to maintain the active card information for my own benefit. How dare you compare me to UB or Absolute Poker? Players want a solution to those problems, we have one and you are obviously too narrow minded to accept that. Why you keep fighting it is not clear, only that you do not have players best interest at heart. I on the other hand am seriously concerned about players getting cheated and I intend to put a stop to it.

    Our technology will not allow anyone at our company or the poker room site to know which cards are being played at specific tables. Our technology keeps the information hidden from everyone (us included) until the end of the hand. Once our invention is patented, the world is free to read how it can be done. At this time, DealGuardian demonstrates what has been described in our patent application and is the only working implementation we know of.

    As for the numbers...

    The probability of one person getting a specific pocket pair at a 9 handed table is: 4.0770%

    After 52,000,000 hands my data produced pocket pairs of: aces - 4.0700%, kings - 4.0746%, queens - 4.0740%, jacks - 4.0721% and tens - 4.0752% of the time.

    The difference between these two numbers are: aces - 0.000074, kings 0.000024, queens - 0.000030, jacks - 0.000049 and tens - 0.000018

    My original number 0.0074% was rounded to the delta between the statistical probability for pocket aces and the results of my data.

    The personal attack I was referring to was some apparent misspelling I did that you continued to mentioned over several posts. You did find one typo in my document (thanks) and it has been corrected.

    There is no real regulation for the online gaming industry. I have spoken with a representative from the Isle of Man and they do not personally inspect the code running on the servers for the sites. Having 3rd party code certification is a plus, but no one is testing all active servers to see if they are running the certified code.

    We already have described our technologies on our site. As I have stated before, we use industry standard practices to secure the active card data and to produce deck shuffles.

    Posted 1 week ago #

  13. eblade
    Moderator

    Sorry I was away for a while, I've been busy with things outside of the Internet. I'd just like to point out that Jim, who is far better at the math side of things than I am, is doing a much better job of explaining exactly what is wrong, even though as a non-math guy, it was obvious to me that the whole thing appears flawed - especially on top of the sideways remarks indicating that you believe the entire industry is cheating, by making high pocket pairs more likely to run against each other, somehow or other. Frankly, if I had a dollar for every time I've dealt AA vs KK or KK vs QQ or AA vs JJ or even TT vs 33 over the year that I worked in a charity room, I'd have more money than a flawed study, a flawed dealing algorithm, and unknown encryption and transport methods is worth.

    So, which patent application is it? You do realise that those are a matter of public record, right?

    Posted 14 hours ago #

  14. nmaiorana
    Member

    There is nothing wrong with our algorithm, it matches what is statistically supposed to happen.

    Our shuffling algorithm is not part of our patent application. We are still in a patent pending state, and we did not make the application public. Yes, we do realize that what we are proposing in the patent will be made public. However, our implementation of the patent we applied for does not need to be made public.

    We are in the process of making the shuffling code Open Source, so look for it at Source Forge soon!

    Posted 5 hours ago #

  15. nmaiorana
    Member

    Just in case you think the Poker Fraud thing is taken care of, here is some news about Pit Bull Poker: http://www.pokernewsdaily.com/pitbull-poker-under-fire-over-security-superuser-accounts-3773/

    Posted 4 hours ago #

  16. eblade
    Moderator

    Have you read the second post in this thread?

    Posted 4 hours ago #

  17. nmaiorana
    Member

    Yes. I and responded by showing the the 4.08 number that I used was rounded up so the difference was actually less than the 0.01% Jim stated.

    --->
    The probability of one person getting a specific pocket pair at a 9 handed table is: 4.0770% (which was rounded to 4.08% in my document)

    ... my data produced pocket pairs of: aces - 4.0700%, kings - 4.0746%, queens - 4.0740%, jacks - 4.0721% and tens - 4.0752% of the time.

    <---

    Think about it this way. The more trials you run, the closer you get to the statistical probability of an event, but you will never hit the exact computed probability. If you only did 100 trials, then your results will be close the the statistical average of the event, but either a bit too high or low. Run 1000 trials, and your number should get even closer. As you keep running more and more trials your number will get close, but may never meet the computed probability.

    Flip a coin a few times and compute the average rounding the result out 4 decimal places or more. Only when you flip enough trials to have the same number of head and tails, then your number will not be 50/50. It will be close, but not exact.

    You asked for our shuffling code and my partners and I agree that those modules are not the main purpose of DealGuardian. So we have decided to Open Source the Playing Card and Shuffling code. It should be out in Source Forge in a couple of weeks.

    DealGuardian is so much more than an electronic card shuffling service. It allows poker rooms to deal games without having the data for insiders to defraud honest players. DealGuardian secures your card data from our servers all the way to your PC. And since DealGuardian maintains the deck information, no one can peek to see which cards are coming.

    This is all done in such a way so that it is impossible for for someone at DealGuardian, even if they work with an insider at the poker room, to determine what cards are being played at specific tables. Yes, this is all spelled out in our patent application. Yes, we use industry standard technologies to operate securely. And Yes, our technology would have prevented the UB, Absolute and now Pit Bull poker scandals.

    Posted 2 hours ago #

Reply

You must log in to post.