Full Tilt on Randomness and Audits

[BaddBeatBobb]

I'll start this thread with a simple posting of a sequence of emails between Marty and various levels of management at FT Poker.

****************************************************
I. Initial Query from Marty

[snip]
More than any other competitor's site, there are concerns and
allegations in regards to the fair distribution of hands at Full Tilt
Poker. I was wondering hence, why doesn't Full Tilt Poker publish a 3rd
party hand distribution audit? (see PokerRoom.com)This seems to me a
reasonable thing to do if at least to alleviate concerns in the poker
community.
[snip]


II. Reply from Full Tilt Front Line Support

[snip]
Unfortunately, I do not have the appropriate information available to
properly address your concern regarding our Random Number Generator
verification. That said, I have forwarded a copy of your email to the
appropriate department who will address your concerns. You should
receive a reply from them in the near future.
[snip]


III. Reply from FTP Upper Management

[snip]
The random number generator used by Full Tilt Poker is very robust,
since it uses several redundant and independent sources of entropy. To
generate a 32-bit random number, 3 different 32-bit random numbers are
generated by independent systems and XOR'ed together. XORing a random
value against any other value yields a value just as random as either of
the preceding values. So even if two out of the three sources were
failing to generate sufficiently random values, the final XOR'ed value
will be random provided at least one of the three numbers is random.
This holds so long as the two values are uncorrelated -- if the values
are correlated, the randomness could be cancelled by the XOR operation,
so under this scheme it is critical that all the different values come
from independent sources.

In summary, we use two different pseudo-RNGs plus a physical source.
The three sources are (1) The ISAAC pseudo-rng, (2) the OpenSSL
pseudo-rng, and (3) a hardware rng that has a physical source of
entropy.

As for a third-party independent audit of our shuffling and dealing
processes, we are currently considering our options in that area and
hope to have a related announcement soon.
[snip]

*******************************************************

So, on an initial analysis, a few things are going on. First, Full Tilt is definitely taking the time to respond, rather than sending out the generic, "Everything is fine, and it's all fair and random" form letter. One line from the first email is rather crucial in explaining this new opennes:

[snip]
We do understand that there have been some security scandals associated
to some of our competitors.
[snip]

FTP really does *not* want any of the taint to stain them, so they're giving a peek behind the curtain, and they're even considering an audit. This is all good news.

Now, as to the question of their random number generator (RNG), I have a few comments. I'm surprised to hear they are using software-based pseudo-RNGs (pRNG). The main problem with pseudo-random is that it's *not* random. There was a poker site that went down in flames a few years back when groups of mathematically skilled players learned which algorithm was being used and were able to anticipate the deal.

Here's an article on pRNG's:

http://en.wikipedia.org/wiki/Pseudorand ... _generator

Now, do I think that all those donkeys are clever programmers who anticipate the deal? Not at all. However, pRNG, if it at all relies on inputs from a current state of conditions to generate the next state, may generate future states which do not differ sufficiently from present states to be equivalent to what chance alone might produce.

Ideally, if you insist on using a pRNG, the sequence of an entire deck would be predetermined by the algorithm before any cards hit the table. However, if due to bad programming or some sort of carelessness, future cards falling off the deck are connected in any way at all to the current situation, trends will begin to appear. If donks with a 30% chance to win based on a physical deck of cards begin to win 31% of the time (or more), over millions of hands, people will begin to notice that donks are winning tens of thousands of times more often than they should.

Because of this, most poker sites now rely entirely on sources of true randomness, such as global temperature variations between various weather stations. Truly random bits can only be generated from physical sources, not algorithms. FTP indicates that their third source of bits is a truly random source (but they didn't say what that was).

Now, the FTP representative did say that the three bit sources are combined using XOR (exclusive or, in common language "either tea or coffee or milk") functions to ensure randomness. However, if you have a single true source of randomness, you don't need to combine anything.

I'm now more anxious than ever to hear whether FTP will submit their shuffler to an independent audit. . .

[JP]

Regarding what you said about the rng at FT, this is what I have been posting for almost a year now. On occasion, I am able to pickup on a string of correlations that give me a feel or "hunch" on what to play. Cards I wouldn't normally play based on "patterns" that show up. WHEN I pick these patterns up, they pay off handsomely.

I am just a regular guy, and if it's THAT blatant that I can see it, it has to be a pretty poor rng. If I KNEW what I was looking at and had the math skills to read it better, I could really use that information to gain an unfair advantage over all the other players. I DO gain an unfair advantage when I see it and use it to win with. FT helps me win their money when they make it that easy for some simple schmuck like me to see it. I don't feel "guilty" about using this "read". It's SO rare and I get slammed by it so often, I figure it averages out.

The reason for the donks winning with junk so often due to this rng isn't because they are math whizzes either. It's simply because they play bad and the variance isn't "normal" or random. They win with bad cards and bad plays, so the think it's the way to play, play that way, do not get better, but still win because it's NOT true random cards. Make sense?

[gadget51]

I think I would agree with both above comments in general. It is true that algorithmic RNGs produce seudo-random results by their very nature. I do wonder though, if FTP omitted the third source so as to preserve it's integrity? Probably, I expect.
I wonder if these pRNGs are the cause of so many 'textured' boards? for example, if one holds, say, 4a5b and the flop is say, 279 rainbow, it just seems a little too common that either one hits their str or the river pairs up the 2; first and last card pairings appear particularly common.
However, I haven't done the math on this so I could easily be talking hot air! I play many hours a day and it just seems that way, so maybe it's just the number of hands I see.
Superb post though and good to see FTP and the like are at least acknowledging there 'may' be a problem.

[Coop]

If donks with a 30% chance to win based on a physical deck of cards begin to win 31% of the time (or more), over millions of hands, people will begin to notice that donks are winning tens of thousands of times more often than they should.

Regarding what you said about the rng at FT, this is what I have been posting for almost a year now. On occasion, I am able to pickup on a string of correlations that give me a feel or "hunch" on what to play. Cards I wouldn't normally play based on "patterns" that show up. WHEN I pick these patterns up, they pay off handsomely.

I am just a regular guy, and if it's THAT blatant that I can see it, it has to be a pretty poor rng. If I KNEW what I was looking at and had the math skills to read it better, I could really use that information to gain an unfair advantage over all the other players. I DO gain an unfair advantage when I see it and use it to win with. FT helps me win their money when they make it that easy for some simple schmuck like me to see it. I don't feel "guilty" about using this "read". It's SO rare and I get slammed by it so often, I figure it averages out.

The reason for the donks winning with junk so often due to this rng isn't because they are math whizzes either. It's simply because they play bad and the variance isn't "normal" or random. They win with bad cards and bad plays, so the think it's the way to play, play that way, do not get better, but still win because it's NOT true random cards. Make sense?

I have observed this since the discussion around this and other matters regarding the secure and safe use of online poker clients has come up here. I have followed British Bulldog's posts about Absolute Poker and read the threads at other forums that were linked.
I have read all the FullTiltPoker-accusations about their RNG and all the bad-beat stories by lVlethod.

Several things come to my mind now as I adress this subject for the first time.

The Absolute Poker subject and the disqualification of the formerly winner "The VOid" in Pokerstars' WCOOP Main Event scared me pretty much, but the way several influential forums took their part in clearing the mess up in the first mentioned case build up my trust in a more secure future of online poker.

The Full Tilt Poker case is by all means different from the two above since it does not involve any kind of cheating. The client itself is being under suspicion of not dealing cards randomly.

While I am now, after having read a bit about pRNG, pretty convinced that FTP's cards are not shuffled properly, I still do not think that there is either any difference between the $2 and $5 STTs (the pRNG is used at both limits!) or that the differences in randomness that occure because of the false RNG could be seen or recognized as patterns by any human without using a computer.

I furthermore think that the following statement is completely wrong:

Bad players/donks win more often because of the pRNG/ cards not being deal completely random.

I think that bad players as well as good players profit and lose because of the pRNG in the same way.

[BaddBeatBobb]

Hi Coop,

Yes, I think you make a good point: *If* the deal of the deck is completely pre-decided before the hand starts, then donks have no special advantage.

My point (lost in my long rambling post) is that if, somehow, the pRNG algorithm depends on present state inputs to generate future outputs, that this is totally unsuitable for a card shuffler, and it could explain the perceived high prevalence of unusual suckouts of a particular kind under particular circumstances.

[Coop]

Ah, I understand now.
But it could also mean that under certain circumstances Aces would hold up more often. So it's still no advantage for the donks, is it?

[BaddBeatBobb]

Quite possibly, but since there is no independent audit, we really have no way of knowing whether the pRNG sources together with the claimed entropic source actually do generate a fair and random deal.

[JP]

The point is, the donks DO win more often because the cards are NOT accurately "random". They are skewed toward that "lesser" variable therefore the worst players that play those cards have a higher potential to catch a winning hand. It is not equal because it does not deal the same potential to players playing top cards or "accepted" strategies considered "proper" poker by most "normal" players.

Normal as in those with a brain, who realize that 95os or 7Tos is NOT a great hand to bet, raise or go all in with preflop in a tourney, with a full table and plenty of players still in the game. These "normal" players in every "accepted" strategy would never play those cards against a big stack if they have to "pay" from any position and go in strong preflop with them. Yet these same dead cards that would say mean something 15% of the time, are meaning something more like 30-50% of the time if anyone wants to risk it all and keep playing junk til the end and hope to win.

By the same token, there are normal strategies and normal players that WOULD push or call strong with 88 preflop. Yet it is not "normal" for AA to go against 88 and lose so often. In fact, AA is NOT normal to lose to every lesser hand it comes across as often as it does on full tilt. No matter WHAT you want to call it, AA should not be the underdog but that is basically what has been happening. It is not just AA either. It is KK, AK, QQ etc... top hands that consistently lose much more often than normal historical percentages to absolute junk or lesser hands.

I also think there is something I read about the FT deal. I may be wrong, but the way I understood how the cards were generated, was that it recalculated after every set of cards was dealt and generated the next cards to come. That is akin to shuffling the deck after dealing hole cards then dealing the flop, and shuffling the deck again.

once cards are shuffled or the "numbers" are generated, then all the cards for the rest of that hand are or should be predetermined for the final outcome just as the cards in a live deck are fixed in place post shuffle. The cards to come are predetermined at the onset of the deal.

I played the identical games and the identical strategies yesterday and today. yesterday, I did not experience many suckouts. Yesterday I was up over $200. Today, I was hit left and right and am down about $60. If it were truly "random" and following basic percentages, my numbers would be less on the win side and less on the loss side. they are skewed because it is not random. I was not rewarded yesterday by getting dealt good cards only. On the contrary, most of the cards were mediocre and few AA and KK hands dealt out to me. I played well, got rid of junk, my "upper hand" stayed upper and wasn't sucked out on as much as the FT "norm". Today, my upper hands did not stay upper.

[BaddBeatBobb]

Does anyone have a sense of whether the alleged skew affects ring games?

I was playing 5c/10c NL cash yesterday. Went down about 25 bucks, mostly to one particular fishcake. Her holecards were just godawful, but she was hitting flops like a ton of bricks, over and over and over. It even got to the point where she was goading me. I kept playing, assuming I'd get my stack back, but no such luck. I just kept draining money. I did have to laugh when my AA got cracked by her 74s, all-in preflop.

Now, maybe she's some uber-genius, and I was being radically outplayed, but it didn't have that kind of feel. It was like being up against the world's luckiest card rack.

JP, if FTP does indeed derive future outputs from current inputs (holecards, flopped cards, etc), then there could indeed be a weird but subtle bug in the randomizer. The funny part is, it would only affect microlimit players, because that's where the donks live. Top level pros would likely have quite different scenarios (inputs) which would then generate different outputs.

[JP]

this is true, which is why so many top players are taken out by bad beats. The pros and solid "normal good" players won't play that crap. In the higher stakes games maybe there is a difference. I did not notice it in games up to $50 that I have played this weekend. Underdogs rule was the song of the day.

[waibeans]

i also think that it affects tournament results alot more than ring games for the reasons that with elimination play you find alot more all-in events. Short stacks on their last legs will only be able to shove and there by the hand will have to be played out till showdown giving more opportunities for bad beats.

I noticed this mostly in playing SNG's ITM or in midgame MTT's where even if stack sizes are similar in size the blinds are enough to make every other hand an all-in situation. With a flawed RNG this is where you see most of the Bad Beats hitting runner runner str8s and flushes as opposed to hitting 2nd pairs or 2 out sets.

very good post tho BBB.. thx for taking the time to look into it!

[gadget51]

whilst bowing to what seems a general concensus of opinion, I would have to say, even given my own scepticism, that my being sucked out on time and time again [as has happened lately for weeks], that such a short run of hand histories would not constitute proof of failure in the system - as much as I'd like to think it. Hence, as as been reiterated by those more knowledgeable than me, the need for a formal audit. I do think that without this, the rest is conjecture.

[BaddBeatBobb]

Quite right gadget. The question now is, will Full Tilt hear the growing chorus of discontent and doubt? Even if the randomization is impeccable, and we're just a bunch of whiners, they could do themselves a lot of good by setting the minds of their paying customers at ease.

[JP]

Here is an example of their sequencer/randomizer being exploitable:

Remember in another posting about card Sequences at FT? I picked up on 6’s at FT just now. Way too often hitting sets and boats, IF I played them and if others did if they had them too because they were hitting the board every other hand and a few times back to back.
So I’m sitting in the button and get 69o, cheap flop as chip leader and limp in. Flop comes and BAM! 6Q9. Rest is history and I take it down. Hand is below. By the way, the two hands before this had 6 on one flop and 66 on the other with the 66 flop immediately preceding this hand.

Full Tilt Poker Game #4080810658: $6 + $0.50 Sit & Go (Turbo) (31009753), Table 1 - 40/80 - No Limit Hold'em - 19:53:40 ET - 2007/11/05
Seat 1: Longtermlucker (1,155)
Seat 2: StokeMoney (1,380)
Seat 3: TY618 (945)
Seat 5: LBSpoker (1,100)
Seat 6: Big Don Vito (1,600)
Seat 7: JPcontender (3,900)
Seat 8: marcolino2001 (2,050)
Seat 9: pugbac (1,370), is sitting out
marcolino2001 posts the small blind of 40
pugbac posts the big blind of 80
The button is in seat #7
*** HOLE CARDS ***
Dealt to JPcontender [6h 9d]
pugbac has returned
Longtermlucker folds
StokeMoney folds
TY618 calls 80
LBSpoker folds
Big Don Vito folds
JPcontender calls 80
marcolino2001 calls 40
pugbac checks
*** FLOP *** [6d Qc 9c]
marcolino2001 checks
pugbac bets 100
TY618 raises to 865, and is all in
JPcontender has 15 seconds left to act
JPcontender calls 865
marcolino2001 folds
pugbac folds
TY618 shows [Qd 3s]
JPcontender shows [6h 9d]
*** TURN *** [6d Qc 9c] [Ts]
*** RIVER *** [6d Qc 9c Ts] [8s]
TY618 shows a pair of Queens
JPcontender shows two pair, Nines and Sixes
JPcontender wins the pot (2,150) with two pair, Nines and Sixes
TY618 stands up

Later on, heads up, I'm in 2nd place, I have A3 suited and raise it up. He calls, flop comes 636, he checks, I raise, turn is a 4, he checks, I go all in and he callse with K4, he wins, but 6's kept hitting until the end. Sure wish I had A6 that time! I had folded A6 earlier this game and same thing, set on the flop if I stayed.

[JP]

Here's another sequence I just now exploited. 5 of the 7 seen flops had 1 or more 5's in it. In this one, I had 55, I thought hmmm, lets check the hand histories quick. One PFR and one caller, I look and come back and raise it to 500, one goes all in, the other folds, I call. He has AK, flop hits and BAM! I get a 5 for a set and take this one down. Hand is below:

Full Tilt Poker Game #4081210948: $6 + $0.50 Sit & Go (Turbo) (31013462), Table 1 - 20/40 - No Limit Hold'em - 20:31:00 ET - 2007/11/05
Seat 1: Gu1ness (1,605)
Seat 2: TreyFu (1,455)
Seat 3: BigBen1919 (1,485)
Seat 4: Fugowi (1,220)
Seat 5: JayA21 (1,520)
Seat 6: Big Don Vito (1,630)
Seat 7: hondagurl (1,440)
Seat 8: Hefpow (1,155)
Seat 9: JPcontender (1,990)
Hefpow posts the small blind of 20
JPcontender posts the big blind of 40
The button is in seat #7
*** HOLE CARDS ***
Dealt to JPcontender [5s 5d]
Gu1ness folds
TreyFu folds
BigBen1919 has 15 seconds left to act
BigBen1919 raises to 140
Fugowi folds
JayA21 folds
Big Don Vito folds
hondagurl folds
Hefpow calls 120
JPcontender raises to 500
BigBen1919 raises to 1,485, and is all in
Hefpow folds
JPcontender calls 985
BigBen1919 shows [Ah Kd]
JPcontender shows [5s 5d]
*** FLOP *** [6s 5h Qc]
*** TURN *** [6s 5h Qc] [Kh]
*** RIVER *** [6s 5h Qc Kh] [7d]
BigBen1919 shows a pair of Kings
JPcontender shows three of a kind, Fives
JPcontender wins the pot (3,110) with three of a kind, Fives
BigBen1919 stands up

I don't do this often and it's not so blatantly obvious as these two games were. But if even lowly ME can spot these trends, think what a braniac can do to you!

Just now in this same 5 game, I have 22 and decide to call the 3xbb pfr. Flop comes and BAM! 5 8 T, he bets, I fold. Next hand? QJ in sb, I limp, flop comes and BAM 45J, I bet, he folds. I'm still in it and 5's are still hittin' it.