How to Spot Bots and Collusion in Online Poker Games

i've been in crypto since 2017 and i've seen every angle. poker bots, collusion rings, mixer abuse, on-chain analysis that catches everything. the problem is, collusion in online poker is getting harder to detect because the actors are getting smarter.

first, let's be clear what we're looking for. bots are automated players. collusion is two or more players working together against others. these are different problems requiring different solutions.

0x01bot detection: what the data shows

bots have tells. they do not variance. they do not tilt. they do not play faster when winning or slower when losing. they have a fixed strategy that they execute every hand. this is simultaneously their strength (perfect execution) and their weakness (perfect statistical predictability).

if you play 1000 hands against a human, you will see significant variance in their hand selection. they limp sometimes, raise sometimes, fold sometimes. the frequencies follow a rough pattern, but there's noise. the noise is human.

a bot plays the same strategy every time. check 1000 hands and you see preflop open-raise frequency of exactly 18.3 percent, consistently. you see cold-call frequency of exactly 6.7 percent. the numbers are too clean.

the harder part: modern bots add artificial variance. they intentionally play suboptimally sometimes. they slow down. they randomize bet sizes. this mimics human play. you cannot catch these just by watching.

what works: data aggregation. run 2000+ hands against the suspect account. pull the statistics. look for patterns that make no sense. if they 3-bet fold to a 4-bet 82 percent of the time, and 3-bet call 18 percent of the time, every single time, across different positions and stack sizes, that's suspicious. humans adjust.

stake, bc.game, cloudbet all have these analytics available if you export your hand histories. use them.

0x02collusion signals

colluders want to play together against you. they make it obvious if you know what to look for.

signal 1: hand-folding patterns. two players fold in sequence before you, allowing one of them to play against you heads-up. if this happens more than 8 percent of the time, odds are 100 to 1 against pure chance. colluders fold weak hands immediately, leaving the strong hand to face you alone.

signal 2: duplicate play against specific opponents. you see the same two players at every table where you're playing. this is possible by chance, but if they're always at the table when you're there, and never when you're not there, that's suspicious.

signal 3: imbalanced aggression. they're hyper-aggressive against you, passive against each other. you win a pot, they're hammering you. one of them wins a pot, the other folds. this is visible in the bet patterns.

signal 4: consistent positional advantage. they always have position on you, or one of them does. this is harder to fake. if you play 200 hands and you're out of position 60 percent of the time against the suspect pair, that's beyond variance.

0x03the mixer problem

mixers are services that anonymize blockchain transactions. you can deposit 5 eth, run it through a mixer, withdraw 5 eth to a different address. the trail is broken. colluders use mixers to hide relationships between accounts.

this is not illegal in crypto. mixers are legal tools. but in the context of casino gaming, they're a red flag. if an account has mixer activity on-chain and plays poker suspiciously, the account is worth flagging.

chain analytics companies like chainalysis can track mixer flows, but they're not foolproof. a sufficiently patient operator can layer transactions, use multiple mixers, and obscure the source. but the effort required is higher than a casual cheater would invest.

0x04what platforms actually do

stake and bc.game run statistical analysis on hand histories. they look for: impossible hand selection patterns, statistical anomalies in aggression, clustering of specific players in tables where another specific player is present.

they also monitor account funding. if you see multiple accounts funded from the same wallet, that's suspicious. if you see accounts funded from addresses that have sent money to known bad actors, that's suspicious.

they flag accounts for manual review. a human being watches actual hand replays. they look for: is this bot playing, or is this just a tight player? are these two people colluding, or did they just both happen to fold and get dealt bad hands?

this is imperfect. some bots are perfect enough to pass review. some collusion rings are sophisticated enough to look random.

0x05provably fair: the nuclear option

some crypto-native casinos are implementing provably-fair poker. this means the deck is hashed before cards are dealt. players can verify, post-hand, that the cards were not modified. this prevents dealer collusion. but it does not prevent player collusion.

player collusion is harder to solve at protocol level because the colluders are not part of the system. they're just using the system. the solution is monitoring and bans.

0x06practical moves

1. export your hand history. if the site supports it, download raw data.
2. run statistical analysis. look for the patterns i mentioned.
3. report suspicious accounts to the operator.
4. play on sites with good abuse detection (stake, cloudbet have decent systems). avoid unknown operators with no verification.
5. play deeper stacks. bots and colluders struggle with game theory optimal play at deep stacks. they shine in short-stack games.
6. play live games if available. harder to bot, and you can observe the dealer and other players for collusion tells.

0x07the meta-level

the platforms have incentive to catch this stuff. collusion and bots drive away good players. if a poker site becomes known for collusion, the games get softer (because smart players leave) but also smaller (because the good players leave). the operator loses revenue.

but the platforms are also fighting a constantly-evolving threat. the moment they patch one detection method, bot developers and collusion rings evolve around it.

if you're playing serious money, play on platforms with strong reputation for abuse prevention. this is not a free market where best practices auto-select. this is an arms race. choose the operator who's winning the arms race.

stay paranoid. trust the data.