Hello NVG,

We have found evidence of what appears to be a bot ring playing in the iPoker network’s small stakes PLO cash games.

We have filed several complaints with the iPoker network, but after 14 days the accounts in question are still active.

As there has been no action taken from the network, we feel it is best to publicise our findings in order to prevent further harm to players in these games.

We have identified 5 accounts with very similar values across a broad range of statistics. This suggests interconnectedness between the accounts in the form of automated play as this is a statistically unlikely scenario.

Our evidence consists of:

• Accounts with nearly matching values across a broad range of statistics.
• A play style that is somewhat unique compared to a control group of confirmed winning human players.
• Seemingly unnatural timings on actions that differ from human winning players.

We suspect that there may be even more bots than these 5 accounts, but these were the easiest to identify based on their similarities.

There are also some indications that the bot ring might be using hole card sharing strategies to increase their edge.

Accounts & Statistics:

HM2 Screenshot with names & preflop statistics:



Screenshots from Pokertableratings.com :
(Player comparison)




Indications of collusion*:
(*Supporting evidence, EV disparity can be explained by variance and performance on different board textures by skill difference.)

All-in on flop graph:
(EV disparity)




Probability:


Performance by Flop texture:

Two tone:
(Filter saw flop, no 3Bet)



This is how regs from other site with similar tight pre-flop tendecies (vpip <23)fare this particular flop texture:
(Same stake, Same filter)



When flop comes paired we get similar results with the suspected group:
(Same filter)



Group of TAGs on other site with paired flop texture:
(Same Stake, Same Filter)




Footage of timing tells:

I was skeptical of these claims at first when you posted cpl weeks ago in plo forum, but the graphs at the end in particular do seem to indicate some type of collusion going on

Bots on Ipoker? Impossible...

Willy can you somehow run a filter on how often the suspected accounts play with the other suspected accounts? It should be almost exclusively with one of the others also at the table if they are colluding. If more (so far unidentified accounts) are involved then it will skew the result of that though.

Wow, so many familiar SNs there... (I'm not a big grinder, though, don't have sufficient samples on any of them.)

I wonder why only one of the suspects (Anark) took a seat at the top right table at the beginning of the video - there was one more seat available, and if he had 'partners in crime', they must have seen it and jumped there... or was it a complicated way of confusing the security team?

Is that bb/100 accurate ? 500bb/100 ?? I know PLO is swingy but 500bb/100 for 80k hands is pretty ridicilous.

Unfortunately I'm not surprised there's bots that collude on Ipoker, and unfortunately, Ipoker also won't do **** about it, except maybe make sure they have a good rakeback deal.

It's filtered for all-ins on the flop not a naked 80k sample of all hands. This is not an unusual number.

Hmm, out of a dozen or two of times I sat with either of them at a table, only once did a pair of them sit together; otherwise I had only one player from this group at the table. Maybe I was lucky or especially good at denying them seats (jumping into them myself before they noticed), but I don't believe they're a team, or at least they haven't been teamed up for long enough and most of the sample is from the time when each of the accounts played on its own.

Once, when I sat down at a table and it suddenly broke at the same time, Anark played 6 hands with me HU (I don't remember who sat out)... is this botty behaviour? I thought HUPLO is such a difficult game that 6-max bots are programmed to sit out except vs a team member. (We hadn't played each other before that, though, he might have regarded me as a presumable drooler.)

Also, could OP hint at what the timing tells are? I'm bad at spotting them.

if someone has a colluding + winning bot and is using it to grind plo50 on ipoker then he is wasting his talent big time.

They take min 3 sec every pre flop decicion. Some of them take closer 5 secs.
I haven't encountered a single snap fold/ call.

However robotic timing is only a small piece of supporting evidence.

If you have winning Poker (PLO) AI you can scale your operation to any profitable stake level.
If a good multitabler can play up to 20 tables how many tables you can set a botnet to play?
As said I have analyzed only small stakes data - I can't say anything about iPoker mid stakes.

I'm not sure what you are trying to say. Are you implying that you don't think the near identical stats are an indicator of automation of play? What would your hypothesis be?

3:50 in video bottom mid table, both "bots" insta-fold to btn raise

edit: 5minute mark upp left table, both "bots" fold quickly to preflop raise. I'm done looking for this now..

He is saying that creating a winning PLO colluding bot would be an immense task and the brains required would be wayyy better spent doing something else.

Also, with slight modifications (interaction with table selection software), such a bot would have a bigger hourly bumhunting at PLO100-200 in addition to its normal PLO50 grind

If they stack tables, they hardly ever can click advance fold buttons, and it usually takes them a few seconds to respond because they often have other action-requesting tables in the queue.

There is more footage that maybe better showcases the timings, but this particular video is of value because it showcases 3 of the 5 accused accounts sitting in one game at the same time.

Some of them 4 table and some of them 6 table so I doubt they play stacked.

My original question still stands. Are you saying that 5 accounts with near identical stats is not an indicator of automated play? Your basically ignoring evidence because you think it would be unfeasible to create a PLO bot even though you have no expertise in that field.

I`ve always been fascinated bout AI and gametheory. I`m quite sure the person that programmed these bots shares this fascination. Coding a bot that beats humans with a significant edge is probably extremely satisfying, making money with it is like a sideproduct.

You learn a lot by doing this so I wouldn`t say it`s wasted talent. Still it`s against the TOS and sharing holecards is straight up wrong. I`ve always wondered about this when I was watching HS-PLO when people who were in the same room were playing on the same table. Knowing 4 extra cards in PLO is huge.

Wait, aren't they supposed to sit in gangs of 3 all the time, not at some cherry-picked moments?

Why don't they fill up vacant seats at the tables in the footage? Why do they play 3-max when they could gang up at a full table?

Did you scan the entire lobby of PLO20-100? And why wouldn't a bot play at more than 6 tables?

You could make that argument for every nosebleed reg today. Still they pursue their career in poker.

There is not huge leap creating NL AI-engine to develop a PLO one.
After all very sophisticated NL bot rings are well documented. What a waste of talent...

PTR buddylist tells what tables they are on.

I don't have deep knowledge how they set their table numbers.
Maybe some botter could answer to the latter one?

1) Similar stats can be also explained by playing according to the same chart and the same postflop guidelines, say, being in the same study group or coached by the same person.
2) You might have selected a group of regs basing on their similar stats and then started seeking evidence of them sitting together. Surely, in such a decently sized player pool as iPoker, you could find 5 ABC regs with similar stats who were all running good. Correlation doesn't imply causality.

No, they are not. The strong argument is that they are automating their play. This is what connects the accounts.

After you have made this connection between the accounts, it is worrisome that they are playing the same tables on a regular basis.

A study group playing with charts producing statistics this uniform is one possibility. My explanation is automated play.
Everyone can form their own opinion which is more likely scenario.
I didn't seek evidence of them sitting together. I also did not cherry pick any account based on their run good.