QQPK Bots: 893 Banned, 11,469 Named, Questions Remain

Ask three different parties how many bots were sitting at QQPK's tournament tables this autumn and you get three very different answers. QQPK says 893, all banned. The anonymous investigators who started this story flagged 2,587. And Vojtech Cervinka, the pro and coach behind the MTT Database Review account, has published a list of 11,469 nicknames he says play like software, which he believes is still incomplete.

Those numbers cannot all be right, and the gap between them is the real story now. Since the original report on Two Plus Two on October 9, QQPK has gone from holding emails to a written apology, 893 bans and a $300,000 compensation pool. It has also offered an explanation for the strangest number in the report, the suspected bots' impossible all-in luck, that a respected coach says does not add up. We went through all of it: QQPK's statements, Cervinka's 21-minute breakdown of how the bots play, his full list, and our own equity calculations. Our QQPK review stays live while QQPK's full report, due on October 17, is still to come.

What QQPK admitted on October 15

QQPK's Risk Control Team posted its findings to the same Two Plus Two thread as the original report. Its headline: a "collusive cheating syndicate" whose accounts pooled their hole cards. The room's ban count rose from the 724 it announced on X on October 13 to 893, made up of 567 accounts from the investigators' list and 326 "internally identified by the risk control team via device and associative leads." The other 2,020 flagged accounts sit on what it calls a "frozen monitor list."

The money is the most concrete part. The $300,000 pool is made up of $64,627 confiscated from the banned accounts and $235,373 of QQPK's own funds, so the room is paying in about 3.6 times what it took from the cheats. Payouts will be weighted by the damage to each affected tournament and player, with the funds "subsequently credited directly to players' wallets." It also apologised in writing: "We sincerely apologize to our players for our risk control system's failure to detect this novel cheating method earlier." It promised to release anonymised hand histories, launched a bounty for outside researchers, and offered the original investigators an award and an advisory role.

On the game itself, QQPK says its random number generator is fair, pointing to its certificate from Gaming Laboratories International, the US testing lab, and that "this review found no system manipulation or backdoor exploits allowing visibility of unexposed hole cards." The complete remediation report "will be published on October 17."

The bot hunter's breakdown

Cervinka is not a random voice on a forum. His MTT Database Review account flagged the bot accounts that were removed from ACR's $12.5 million guaranteed Venom in 2024, as Poker.org reported at the time. On October 11 he published a 21-minute video, QQPK Bots Masterclass, built on roughly 25 million hands from about 4 million hand histories, a sample he stresses is "the whole population of everybody," not a filtered group.

His starting point is a habit almost nobody has: a 3-bet to exactly 11 big blinds at 25 to 60 blinds deep. Filter the archive for players with at least 1,000 hands who use that sizing regularly, he says, and on a normal site you would expect perhaps 50,000 to 100,000 hands. On QQPK he got 9.8 million of the 25 million, or 39%. Looking at the same limping tell across the whole population rather than his filtered group, he found almost twice as many of the tell-tale hands, enough, he says, to suggest that "more than half of the hands played potentially were by bots."

The rigidity is the giveaway. The same narrow range turns up in every limp-raise spot, multiway or heads-up, shallow or deep, with "400 aces" and no king-queen at all. Flop bets tend to be exact round numbers, two, three or four big blinds. "I think you could probably train a pigeon" to spot them, he says, adding: "I feel kind of insulted, to be honest." He says no other bot he has studied "was ever as simple to spot after a few hands." For anyone still playing there, his exploit advice follows the numbers: probe and stab relentlessly when they check, iso-raise their small-blind limps (they limp-fold 69% of the time), and treat a big out-of-position river bet as an invitation to raise with anything that cannot win at showdown.

He also built the opposite group as a control: players with at least 1,000 hands who never use the 11-blind 3-bet and almost never limp. On a normal site, he says, that filter produces winners. On QQPK, across 1.1 to 1.9 million hands, "they are getting wrecked somehow."

11,469 names, and five that match

On October 15 Cervinka published the list itself as a public GitHub file: 11,469 unique nicknames from QQPK tournaments between July 16 and October 11, 2026, selected on "very strict, behavior-based criteria." He calls it incomplete, with "at least several hundred additional suspicious nicknames" left off for lack of hands, and he is careful about what it is: unique nicknames rather than confirmed accounts, since name changes can inflate the count.

The names themselves look mass-produced. His notes point to 135 starting with "Happy," 46 of them "Happy" plus eight hexadecimal characters; 245 that are five letters followed by nine digits; a complete grid of seven Thai first names each paired with the same ten poker words; 80 Chinese names built from one template (a surname character, then a fish, tiger, dragon or king character, then a capital letter); and English mix-and-match pairs like RinggitStriker, FeltedGhost and PolarizedBossku. He calls these "circumstantial evidence of bulk account creation," not proof on their own.

We downloaded and parsed the file. It holds exactly 11,469 unique entries, and his template counts check out. About 70% of the names are plain Latin characters (8,006), about 25% use Chinese characters (2,907), 328 are in Thai and 140 in Vietnamese. Then we ran a test of our own. The original anonymous report named five accounts as examples of its suspected bots. All five are on Cervinka's list.

That match is worth less than it looks, and we want to be clear why. Cervinka's sample, roughly 25 million hands from about 4 million hand histories, is almost exactly the size of the anonymous report's (25.4 million hands from 4.18 million histories). Both describe the same limp-reraise range of nines and better plus big aces. They may well be working from the same datamined archive, so the overlap could reflect shared data rather than independent confirmation. What it does show is that two analyses, using different filters and date windows, landed on the same accounts.

The all-in luck: three explanations, one dataset

The original report's most explosive claim was that the suspected bots won far more of their all-ins than their cards entitled them to, a 14.6 standard deviation gap. Playing like a bot does not do that. Something else has to. There are now three candidate explanations on the table, and they are not mutually exclusive.

One: card-sharing. This is QQPK's answer, which it calls the "Blocker Filtering Effect": "Through card sharing, the syndicate accounts knew the distribution of a large volume of dead cards." Knowing what teammates folded tells you which outs are gone, so you go all in when the hidden cards help and pass when they do not. yegor, a Two Plus Two poster with a Verified Coach badge and a forum history going back to 2008, pushed back on October 15: "The card-sharing explanation doesn't match your own EV definition." Blocker effects, he argued, move equity "by fractions of a percent, nowhere near a sustained multi-sigma surplus of actual over EV." That is checkable, so we checked it.

On size, the coach is wrong: known dead cards move preflop odds by several points, and filtering spots on them can match the report's gap with room to spare. On principle, he has a point. Card-sharing only shows up as "actual beats EV" if the EV calculation leaves the folded cards out, the way trackers and the investigators' tools do. QQPK saw every folded hand. If its own calculation counted them, blocker filtering is already baked into the expected value and cannot be the explanation. Its methodology note says only that EV is calculated "only when an all-in occurs preflop, all active players table their cards," which does not settle it.

Two: survivor bias. Cervinka himself does not think the luck is all it seems. He says he has not examined collusion, and points out that any filter requiring 1,000-plus hands favours accounts that ran well: "if they run bad, they are a little bit less likely to pass the 1,000 hand threshold." That is the same selection problem we flagged in the original report. It can inflate a statistic. It cannot by itself explain a syndicate QQPK has now confirmed.

Three: something touching the deck. On October 12, the day before the first bans, yegor described what he says he found in QQPK's client: the whole encrypted deck sent at the deal with the keys released by the server, a claim that "in current implementation privileged users may receive secret keys before the hand ends," and a hidden menu labelled with the Chinese term for a card dealer, which he translates as "manual deal," linked to a production endpoint that answered "unauthorized" when he tested it. None of that is verified, and an "unauthorized" reply proves only that an endpoint exists and is locked. QQPK's denial covers "visibility of unexposed hole cards," which is a different question from seeing or setting the board.

The data fight sits underneath all three. QQPK says the investigators' 4,866,164 raw lines boil down to 1,757,070 unique hands, because datamined archives catch the same hand from several observers, and that it ran 24,026 tournaments and about 22.18 million hands in the window, fewer than the report counted. Duplication inflates certainty by the square root of the duplication rate: 2.77 times duplication (4,866,164 / 1,757,070) would cut the 14.6 SD to roughly 8.8, still far beyond chance. QQPK says that on clean data, "the recalculated EV deviation for the suspected accounts remained higher than the theoretical value," while the 2,020 accounts it did not ban "track closely together, showing no sustained EV deviation." Only QQPK holds the data to separate the three explanations, by running EV for the banned accounts with folded cards counted. If the gap disappears, it was card-sharing. If it survives, someone needs to explain the deck.

What QQPK still has to answer

The 11,469 figure is the hardest for QQPK. If even a large share of those names are bots, the 893 it banned are the visible tip, and the 2,020 accounts it cleared to "monitor" status, every one of which the investigators flagged, deserve a second look. Cervinka's estimate that bots may have played more than half of all hands in his sample would, if anywhere near right, make QQPK's tournament fields a different product from the one players thought they were buying into.

Where we stand

Credit where it is due. Within a week of a forum report, QQPK froze withdrawals, confirmed the cheating, banned hundreds of accounts, apologised in writing and put more of its own money into a compensation pool than it took from the cheats. That matters more than usual here, because its licence comes from Anjouan, a jurisdiction where, as an ABC News investigation reported in December 2025, the Comoros central bank has called several licensing bodies "fictitious structures." In practice, QQPK's own reporting is the main accountability players have.

It also means a syndicate ran on QQPK's tournament tables for at least three months and was first exposed by outsiders, and two outside analyses now put the bot problem well above what has been banned. Our QQPK review stays live for now. On October 17 we will be looking for real payouts, a clear answer on how EV was calculated, a response to the deck-key claims, and some account of the 11,469 names. We will update this article and the review when the report lands.

Sources: Two Plus Two investigation thread (October 9 to 15, 2026, including QQPK Official's statements and yegor's posts); QQPK on X (@QQPKOFFICIAL, October 13, 2026); QQPK Game Integrity Investigation report; MTT Database Review, QQPK Bots Masterclass (October 11, 2026); MTT Database Review's list of 11,469 nicknames (October 15, 2026); Poker.org on the 2024 ACR Venom bots; GLI certificate for the QQPK RNG; ABC News on Anjouan licences. The list analysis, blocker-test equities and deduplication estimate are our own calculations.