Suspected cheating on AP

[citanul]

TT,

I'd be glad to do a little talking about the job thus far (even the poker related stuff), but I have a huge favor to ask anyone who has actually read all this nonsense and kept abreast of the situation:

Steve just asked me if anyone had looked over a large sample of this guy's hands to *actually* see if he was cheating, and in the blatant way that everyone says. Specifically he's looking for someone who has shown some stat breakdown that says something like:

Big pile of hands won money at showdown % 100% or whatever.

Basically it's one thing if the dude just calls a ton on the river and so he wins frequently with bizarro hands, but it's another thing if he always value bets the worst hand, always raises when last to act wit the best hand, and never calls with a loser.

I'm sure that someone has done something like this, but with the sheer volume of threads, I don't even really know where to start (though obviously there's some good options for where to start).

-----------

So our poker research at the moment:

We're working with a site that is providing us with anonymized hands (10s of millions), including every hand that was ever played by a subset of their players. The first thing we're doing is an analysis of results, to show that poker is in fact a skill game. (We're not especially interested in this, but the site is.) Then we were going to look at more interesting things, looking at what sorts of skills people actually have, what things actually work, etc, then we're going to also look at collusion, from STTs to cash games. And a few other things, of course.

[Dids]

Cit,

The infinite river agression is a giant clue imo. For me, that was the "holy [censored] this is real" stat.

He never, ever, ever, calls on the river. Raise or fold.

[adanthar]

It's in the OP in the last thread, dan. He's called on the river *once* in several hundred hands, with bottom pair (that was Q high no draw on the turn - yes, he called a bet there) that lost to a backdoor flush (that was 8 high on the turn). There's a very good chance he just didn't see that flush hit. Every other hand he's played was either a raise or fold - he'd sometimes minbet bluff a worse hand when checked to (yeah, I dunno) or checked a hopeless bluff behind, but he never called a bet with a worse hand other than the Q2.

[adanthar]

oh, and IM me with some info on your research if you can? this sounds fun.

[citanul]

adanthar,

I actually almost called you from work on my way back from a talk, so yeah, I'll be in touch probably soon. Sometime after Yom Kippur is over though. Have an easy fast. (That goes for all you Jews out there, and anyone else who is just fasting for the hell of it.)

There's obviously a bunch I can't say about our research, and on top of that, things are still getting off the ground, but it's all pretty cool stuff.

c

[Leader]

[ QUOTE ]
Cit,

The infinite river agression is a giant clue imo. For me, that was the "holy [censored] this is real" stat.

He never, ever, ever, calls on the river. Raise or fold.

[/ QUOTE ]

It's pretty easy to do a statistical analysis of this too.

Basically we just test the hypothesis that villain in this case is really not cheating, he's just running really weird hot. Assume that normally he has a river AF of 4, still really agro but at least reasonable. This means that 20% of the time he doesn't fold, he calls. Let's say he saw the river and bet or raised 30 times in a row excluding folds. The likelihood of this happening to someone who really calls 20% is .8^30 or .0012. If he calls once over that span it's 30*.2*.8^29. More then 2 is more complicated. Off the top of my head, I believe it's:

(sample size)!/((times he called)!*(times he raised/bet)!)(probability of a call)^number of calls*(probability of bet/raise)^number of bets/raises.

Note that "!" indicates a factorial. 1! = 1 2! = 2*1 3!=3*2*1 and so on. 30! may seem hard to deal with but in this case it's always divided by other factorials. For example 30!/28! is just 30*29.

[adanthar]

Hey Schneids,

If you're still on this thing, here's some very solid circumstantial evidence that it's an inside job.