Last week, a group of a dozen related PokerStars accounts were found to be bots, or automated computer programs that can play poker with limited human interaction. This would not normally be big news, but the fact that the bots were first brought to light by parties outside of PokerStars got people talking.
On July 13th, a member of the Two Plus Two poker forum named “malloc” started a thread, detailing his suspicions about three players whom he believed had playing styles and win rates that were too similar to be a coincidence. He posted some statistics on the players, pulling the figures from the data mining site PokerTableRatings.com (PTR). A poster named “Gugel” took it further on July 16th, finding more statistics that convinced most readers of the thread that the three players in question were, in fact, bots.
But the hammer came down later that same day when the folks at PTR revealed that they had run with malloc’s initial findings and discovered a total of ten accounts — 7emenov, bakabar, crazier, mvra, nakseon, kozzin, demidou, koldan, Daergy, and feidmanis — on PokerStars that they believed without a doubt were bots.
The first thing PTR pointed to was the statistical similarity between all the accounts. With very little variation, they had the same VP$IP (voluntarily put money into the pot) statistics from every table position, the same pre-flop raise percentages from every seat, and the same three-bet frequency from every position. They also folded, called, raised, and bet with the same frequency on the flop. Of all the stats, the most glaring is the flop check-raise percentage: zeroes across the board except for one account that was at 0.1 percent and one that was at 0.2 percent. Even the nittiest, most passive of players check-raises on the flop on occasion.
Taking it a step further, PTR determine the Euclidian distance between one of the bots and all other players who played at least 100,000 hands at $.25/$.50 No-Limit Hold’em in 2010 using six stats. In short, the Euclidian distance shows how similar two accounts are, with the accounts being more similar the closer the result is to zero. The smallest Euclidian distance between the baseline bot account and the rest of the non-suspected players was 32.94, and most were 45 to 85 range. The largest Euclidian distance between the baseline bot and the other nine examined bots was 2.39.
Next, PTR looked at how the ten accounts changed stakes, something that malloc also noticed. They all started at $1/$2 No-Limit Hold’em and moved down to $.5/$1 at about the same time (one account moved two weeks earlier than the rest). They then all moved down to $.25/$.50 within days of each other. Finally, they all stopped playing on either July 15th or July 16th.
The last thing PTR saw was that all ten accounts displayed the same unique betting pattern. Of note, whenever one of bots was in the big blind, nobody raised pre-flop, and the action checked around to it on the flop, it would move all-in on the flop with a good hand. With nothing, it would bet and then fold to a raise. According to PTR, the bots took this line 99.6 percent of the time.
The accounts were banned by PokerStars on July 17th, the day after PTR published its findings. PokerStars claimed that it had also discovered the bots – before PTR, in fact – but was still determining whether or not there were more related accounts before shutting them down. The poker room felt that if it closed the accounts prematurely, any related bot operators would change their programs to avoid detection. As it turned out, PokerStars did find two more bots in the same ring and shut them down on July 19th.
According to PokerStars, the bots cashed out a total of $223,150. $74,439 of that was from cash game profits, $6,305 was from VIP freerolls, and $142,406 was from VIP rewards. PokerStars confiscated $2,084 in cash and $4,446 in unspent FPP’s.

This site is registered on wpml.org as a development site.