Statistics takes down poker cheats

This story of busting a $15m poker scam by and Australian is fantastic – and is a wonderful demonstration of the benefits of being open with your customers. The full story is on poker community site twoplustwo, and has been going on for a while.

Customers of Canadian tribe owned AbsolutePoker and UltimateBet analysed poker hand history, and discovered that some people were winning far more than statistically probable. They raised it with the company (same parent co) and after, apparently, an unreasonably long length of time they agreed that sure enough, insiders had been playing – and winning – hands with the ability to see what all of the other players were holding. Turns out the server software had security holes in it – something that in an open sourced piece of software would have long been plugged.

The cheats were a bit naive, and should have learned from the professionals – if you have a system that can beat the house, then don’t beat the house up (even if you are the house, which in this case it appears they may well have been.). By winning too often it was blatantly obvious that something was amiss. Moreover apparently some of their play was only logical if they could see everyone else’s cards, which raised suspicions of fellow players..

Far better, if you are a cheat and don’t mind the heightened risk of losing your friends, reputation, teeth and/or liberty, to only look at the opposition cards now and then – and thus avoid the temptation of winning too much. Even better, learning from the Blackjack card counters, lose consistently when the stakes are small, and then cheat only when you have gone big and dragged others with you. That may mean losing when you could have won, just to keep the win/loss stats down.

Even these approaches are testable if the data is released. First – look for overall biggest dollar winners, and go after them with a fine statistical comb. Look for anomalies in win/losses for certain types of hands versus the size of stake and what other players have done.

The article does not say whether Michael Josem, the chap that did some of the original analysis, captured the hand data himself or was able to download it. However the Poker community page shows that they got the data through the players collaborating after the fact, and via a spreadsheet that contained a wee bit too much data. Actually the Stuff article misses a lot.

I hope that other poker sites learn from this, and publish their game records for all to see, along with statistics. Meanwhile releasing the source code of the software to highlight any holes wou;ldn’t be a bad thing either.