Cliffs
AP Casino's keno game is rigged. Its RNG is not at all random. Watch our YouTube video to see it in practice.
The Gist
Absolute Poker has this browser-based Keno client (called "Traditional Keno", found under the other games tab
here). It is blatantly not run from a real RNG or even a decent pseudoRNG, but rather from some kind of number generator that produces obvious patterns.
We have no clue if this is the same "R"NG that they use for poker or blackjack or anything else. That seems sort of unlikely because one would hope that someone would've noticed that by now, but given that Cereus ran without encryption for years without anybody noticing, we don't think it would be incredibly surprising to learn that they had an easily crackable "R"NG. So,
we would not play any games on Absolute Poker or Ultimate Bet (they're essentially the same) until Cereus clarifies.
There's some reason to believe that this was not done intentionally but rather was simply caused by some extreme combination of laziness and incompetence. In particular, it appears that AP's Keno "R"NG actually leads to better results for the customer. The customer still loses, and because it's not random, the customer has a much much lower chance of profiting, but it does appear that the customer actually loses less than he would to a fair Keno game.
Regardless of whether this "R"NG is used elsewhere or whether this was done intentionally, it still shows that Cereus does not take the fairness of its games seriously enough. After the
UltimateBet and
Absolute superuser scandals,
the encryption debacle,
the pot that got shipped the wrong way, and the numerous other security problems that players have discovered on Cereus and their repeated promises to improve player security, this is just completely ridiculous.
The Quick Demonstration That It's Rigged
It's a bit hard to explain in words what's going on, especially since nobody plays Keno. This video should make it clear:
(Click image to view)
More Formal Demonstration
(This is going to be a little less rigorous than I'd like because I think what's going on is fairly obvious, and I thought it was more important to just get it out than to do it perfectly. I'm really tempted to add parentheses after like every sentence in this part explaining that I glossed over some stuff, but suffice it to say that something's up and this explanation is good enough.)
Earlier tonight, I was contacted by a mid/high stakes regular who said that a buddy of his told him that AP Keno wasn't random. He described to me exactly what I've shown you guys and proceeded to play a bunch of Keno while I watched him almost always predict when he would hit. I then confirmed it myself and showed it to Thomas (from
our Cake investigation). So, I didn't notice this pattern while I was looking for patterns, I was told what the pattern would be and shown exactly what I was told to expect, and then I confirmed it myself.
In order to confirm it, I decided beforehand that I would run fifty trials of the five-number play ten thing that I showed in the video. I'd record in sequence what "rounds" I hit. I'd then calculate the difference between the first and second round that hit. I'd then count how often different numbers came up and compare the distribution with the expected distribution. Here's that data:
As you can see, five comes up much more often than one would expect. The odds of getting three out of five on any individual round are Choose(5,3)*Choose(75,17)/Choose(80,20) = ~8.4%. So, if the first round that hits is between one and five, the odds that there will be a second round that hits that is five rounds later are the odds that four in a row do not hit and then the next one hits, i.e. (1-.084)^4*.084 = ~5.9%. If rounds one through five do not hit (which should happen (1-.084)^5 = 64.5% of the time, but only happened 4% of the time in my sample), then the next round to hit cannot be five after since there is no such round, so the actual probability should be much lower than 5.9%. However, even if the odds were 5.9%, the odds of something with 5.9% probability actually happening at least 43 times out of 50 are about one in 10^45. So, this is not random.
Other facts about the data are similarly surprising. For example, the fact that one through four never occur. In all fifty trials, the first trial to hit was never later than the sixth rounds, so the odds of the second hit being one through four rounds after the first in an individual trial are 1-(1-.084)^4 = ~29.6%. The odds of never hitting a 29.6% chance in 50 tries are about one in 42 million. I also never saw four or five of my five numbers get chosen. I ran way more than 50 sets of ten and never saw four or five of my numbers chosen, but since I didn't record those, I'll only consider the fifty sets of ten. The odds of getting at least four out of five chosen are Choose(5,4)*Choose(75.16)/Choose(80,20) + Choose(75,15)/Choose(80,20) = ~1.27%. The odds of not seeing this once in 500 trials are about one in 608. I can keep going (for example, in thirty of my fifty trials, both round five and round ten were winners...), but I think the point is clear.
I also recorded some other data. The guy who originally contacted me said that the pattern applies across play tens as well. In other words, if I run one play ten and the eighth round is a hit, then I should expect the third round to be a hit quite often in the next round, since that happens five after the previous hit. So, I crunched the same numbers except now I counted the number of rounds from the last hit of one play ten to the second hit of the next play ten. Here's that data:
(The expected %s are calculated correctly as (1-.084)^(N-1)*.084, so these aren't approximate at all like the above numbers )
So, as you can see, a difference of five occurred 29 out of 49 times when it should be expected to occur only 5.91% of the time. The odds of this happening randomly are about one in 5 * 10^23. So that's not random either.
If I choose a different 5 numbers, the same thing happens. If I choose 6 numbers instead of 5, a different but similar pattern occurs.
So, their random number generator is flawed in some way. We would like an explanation from Cereus as to what exactly the flaw is and how it got into their keno game. We'd also like to know whether this "R"NG was used for any other games that Cereus spreads. Given Cereus's history, we think it's reasonable to expect not only answers to these questions, but also some evidence to support those answers.
(Again, we think it prudent to note that the game doesn't seem to be quite "rigged" in the house's favor. I actually ran better than expected in the controlled trial that I ran.)
Last edited by Kevmath; 10-26-2010 at 04:08 PM.