Cheating in Online Casinos

[Cacarulo]

Hi everyone,


Has anyone ever conducted statistical tests by taking random samples from multiple shoes to check for possible manipulation of the cards?
Or better yet, has anyone actually taken the time to design a program that can detect whether a casino is cheating?

I’m aware of certain methods that could significantly increase the house edge while staying virtually undetectable.
Of course, I won’t go into detail about what those methods are and I’m not accusing any casino of cheating.
I’m simply curious if anyone has made the effort to ensure they aren’t being scammed.


Sincerely,
Cac

[OnlineAP]

I have thought about this a lot, and I do have a lot of online hands that I downloaded that I could put through a gaffe detector. There are suites of tests for randomness of bit strings or cards.

However, I realized that it would be easy to create a program that looks for opportunities to substitute a card that harms the player but actually pushes the statistical properties closer to random. It will take you a very long time to confirm clever cheating.

Very greedy cheaters will be exposed by looking for loss rates outside of normal variance.

Detailed analysis of card correlation and statistical properties will only catch unsophisticated cheating by people who are not too greedy. For example, if they only cheat in certain situations, it will be easier to discover statistically. If they cheat infrequently in a lot of different situations, then it will be very difficult to detect.

[judgeholdem1848]

However, I realized that it would be easy to create a program that looks for opportunities to substitute a card that harms the player but actually pushes the statistical properties closer to random. It will take you a very long time to confirm clever cheating.

I used to always tell people that skilled poker cheats are by definition undetectable. A sharp player won't catch them. Fish don't stand a chance.

[Archvaldor]

Hi everyone,


Has anyone ever conducted statistical tests by taking random samples from multiple shoes to check for possible manipulation of the cards?
Or better yet, has anyone actually taken the time to design a program that can detect whether a casino is cheating?

I’m aware of certain methods that could significantly increase the house edge while staying virtually undetectable.
Of course, I won’t go into detail about what those methods are and I’m not accusing any casino of cheating.
I’m simply curious if anyone has made the effort to ensure they aren’t being scammed.


Sincerely,
Cac

I'm assuming you are talking about live dealer games rather than digital blackjack. There were some attempts to quantify the extent of any cheating in digital blackjack by the Wizard Of Odds and various other people in the early days: for various reasons no one ever really found an effective way to do it.

I am unaware of any large sample size that has been recorded concerning live dealer games for these purposes. There are so many potential biases that could occur I would have little confidence in any such sample. Note that in some verifiable cases cheating was directed at specific individuals.

When you do encounter cheating online it is never subtle-it is usually very obvious. I don't think this is likely to be a terribly good use of your fantastic intellect.

Most online AP's experience overall results within expected statistical parameters. I haven't tracked my overall career results but my monthly results were almost always very much in line with my expected value.

[Gronbog]

Most online AP's experience overall results within expected statistical parameters.

This has been my experience.

[OnlineAP]

I used to always tell people that skilled poker cheats are by definition undetectable. A sharp player won't catch them. Fish don't stand a chance.

Card mechanics in pitched games may be visually undetectable without a high speed camera. I know little about this subject. Cheating for shoes that are shuffled out of sight is not visually detectable. Dealing seconds from a shoe may not be detectable with the resolution of the online cameras.

But whether we are talking about RNG dealt cards or shoe games, the cheating has to leave statistical residuals that can be detected. The RTP is one statistical residual, so unless the cheating creates some residuals, there is no point to the cheating.

Archvaldor's comment mirrors mine. Ad hoc cheating by humans is usually unsophisticated. If you are only cheating to bust the player or prevent the dealer from busting, you make it much easier to detect. You could just as easily cheat by giving the player slightly poorer starting cards and giving the dealer slightly fewer stiff up cards. Then, you could deal all draw cards completely randomly and still have an advantage. Any analysis that just looks at the frequencies of outcomes for each dealer up card will miss the cheating entirely.

My primary approach would look at every single card dealt starting with the first player card. Fundamentally, every card that is dealt changes the EV for the player. I would create bins with ranges of EV delta, dEV. Then I would compare the frequencies within each bin to the population of each rank remaining in the shoe. If an eight is the worst possible card for dEV in a certain situation and a 7 yields the best dEV, do these cards come out of the shoe with the expected frequency. If the outcomes are tilted towards -dEV, you have prima facia evidence for cheating.

This will reveal cheating much faster than using generic statistics like blind card frequencies or correlation functions within the stream of dealt cards.

[Cacarulo]

I'm assuming you are talking about live dealer games rather than digital blackjack. There were some attempts to quantify the extent of any cheating in digital blackjack by the Wizard Of Odds and various other people in the early days: for various reasons no one ever really found an effective way to do it.

I am unaware of any large sample size that has been recorded concerning live dealer games for these purposes. There are so many potential biases that could occur I would have little confidence in any such sample. Note that in some verifiable cases cheating was directed at specific individuals.

When you do encounter cheating online it is never subtle-it is usually very obvious. I don't think this is likely to be a terribly good use of your fantastic intellect.

Most online AP's experience overall results within expected statistical parameters. I haven't tracked my overall career results but my monthly results were almost always very much in line with my expected value.

Yes, I’m talking about live dealer games where the cards in the shoe can’t be seen from the start. The question is simple: how can we know if the shoe has been tampered with?
If the change is large, we can detect it by testing random samples using methods like the Chi-squared test or the Kolmogorov-Smirnov test. But if the change is small,
it becomes much more difficult because there will be too many false positives. This kind of cheating can affect any player.

Sincerely,
Cac

[judgeholdem1848]

But whether we are talking about RNG dealt cards or shoe games, the cheating has to leave statistical residuals that can be detected. The RTP is one statistical residual, so unless the cheating creates some residuals, there is no point to the cheating.

I was thinking more about Russ Hamilton type big-money online cheating, which in various forms was still a major issue when I played. He failed to control his information footprint to keep any sample reaching statistical significance in the aggregate siloed among many players so that the only way they could discern the cheating with any confidence would be through collaboration.

That's eventually what happened. But it was only because as I recall he was pushing 3 and 4 sigma sessions against single players all the time. If he would have kept it .5-1 sigma then bounced, and done that 2,000 times instead of going for a few big scores, no one outside Vegas would know his name imo. He still could have stolen millions.

It appears per your post detecting any kind of superusing or card peeking can be doubly hard because in games where two players have complete strategic freedom and the deck is unscathed, it would be very hard to prove that any play pattern falling within normal population bounds is the result of cheating unless the cheater is blowing you out of pots or hero folding and showing every other hand.

[OnlineAP]

Yes, I’m talking about live dealer games where the cards in the shoe can’t be seen from the start. The question is simple: how can we know if the shoe has been tampered with?
If the change is large, we can detect it by testing random samples using methods like the Chi-squared test or the Kolmogorov-Smirnov test. But if the change is small,
it becomes much more difficult because there will be too many false positives. This kind of cheating can affect any player.

Sincerely,
Cac

If you are restricting the discussion to stacking a physical shoe, then most of my previous comments are not applicable. I was thinking about the case where a physical dealer was dealing seconds or a software program was not choosing the next card randomly.

There are several obvious ways of rigging the shoe.

1) Removing cards from the shoe. This leaves physical evidence that is damning if the regulators are checking shoes after they are taken out of play.

2) Moving cards beyond the cut card to make the cards actually in play be very unfavorable. This is easily detectable by simple counting the ranks of cards in play, but at least the shoe is whole.

3) Stacking the front of the shoe to create a small number of bad outcomes right off the top. If the person stacking the deck knew exactly how many players they are dealing to, this is easy. If players come and go, then the stacking quickly becomes ineffective after a few hands are played out of the shoe. Even if the number players are known when the deck is stacked, online players make many strategy mistakes that would change the flow of the cards that was intended by the stacker. The exception to this is the many forms of infinite blackjack where only one spot is dealt. Many players still split tens incorrectly or hit in double situations. If the deck is stacked to prevent these situations from arising early on, the house can prearrange a very bad run for the players that will be preserved deep into the shoe regardless of the presence of players prone to strategy mistakes.

4) Stacking by creating non-random correlation between nearby cards in the shoe. In this case, the number of cards played in each rank will be exactly the same as for a random shoe. This strategy will be effective throughout the shoe no matter how many spots are being played or misplayed. I don't know how much the house can add to the house edge through this sort of stacking. To gain the maximum advantage, the local correlation will have to be very strong and should be easily detected.

So which of these sorts of shoe alterations were you thinking of in your OP?

I assume it is possible to purchase machines capable of arranging the cards in a shoe in any particular order.

Has anyone ever run a contest to pit stacking algorithms against detection algorithms?

[Cacarulo]

If you are restricting the discussion to stacking a physical shoe, then most of my previous comments are not applicable. I was thinking about the case where a physical dealer was dealing seconds or a software program was not choosing the next card randomly.

There are several obvious ways of rigging the shoe.

1) Removing cards from the shoe. This leaves physical evidence that is damning if the regulators are checking shoes after they are taken out of play.

2) Moving cards beyond the cut card to make the cards actually in play be very unfavorable. This is easily detectable by simple counting the ranks of cards in play, but at least the shoe is whole.

3) Stacking the front of the shoe to create a small number of bad outcomes right off the top. If the person stacking the deck knew exactly how many players they are dealing to, this is easy. If players come and go, then the stacking quickly becomes ineffective after a few hands are played out of the shoe. Even if the number players are known when the deck is stacked, online players make many strategy mistakes that would change the flow of the cards that was intended by the stacker. The exception to this is the many forms of infinite blackjack where only one spot is dealt. Many players still split tens incorrectly or hit in double situations. If the deck is stacked to prevent these situations from arising early on, the house can prearrange a very bad run for the players that will be preserved deep into the shoe regardless of the presence of players prone to strategy mistakes.

4) Stacking by creating non-random correlation between nearby cards in the shoe. In this case, the number of cards played in each rank will be exactly the same as for a random shoe. This strategy will be effective throughout the shoe no matter how many spots are being played or misplayed. I don't know how much the house can add to the house edge through this sort of stacking. To gain the maximum advantage, the local correlation will have to be very strong and should be easily detected.

So which of these sorts of shoe alterations were you thinking of in your OP?

I assume it is possible to purchase machines capable of arranging the cards in a shoe in any particular order.

Has anyone ever run a contest to pit stacking algorithms against detection algorithms?

Hi OnlineAP,

The manipulation I'm talking about keeps the total number of cards in the shoe at 312 while altering the proportion of certain cards. For example, the house edge for a fair game of 6 decks, S17, DOA, DAS, SPA1, SPL3, NS is 0.4059% if players use a total-dependent basic strategy (TD-BS). In this scenario, by simply changing 2 cards in the shoe, the house edge increases to 0.6205%. If instead of 2, 4 cards are changed, the edge rises to 0.8365%, and if 6 cards are changed, the edge goes up to 1.0539%.

That’s why it’s very important to be able to analyze a quantity m of samples, each with a size n, that allows us to detect these changes and report these casinos if cheating is confirmed.

It’s also important for a player to know that a certain casino with live dealers is not doing anything dishonest.


Sincerely,
Cac

[OnlineAP]

Hi OnlineAP,

The manipulation I'm talking about keeps the total number of cards in the shoe at 312 while altering the proportion of certain cards. For example, the house edge for a fair game of 6 decks, S17, DOA, DAS, SPA1, SPL3, NS is 0.4059% if players use a total-dependent basic strategy (TD-BS). In this scenario, by simply changing 2 cards in the shoe, the house edge increases to 0.6205%. If instead of 2, 4 cards are changed, the edge rises to 0.8365%, and if 6 cards are changed, the edge goes up to 1.0539%.

That’s why it’s very important to be able to analyze a quantity m of samples, each with a size n, that allows us to detect these changes and report these casinos if cheating is confirmed.

It’s also important for a player to know that a certain casino with live dealers is not doing anything dishonest.


Sincerely,
Cac

When I play online versus live dealers, the casino always has an edge much greater than TD-BS. Why would they need to cheat by using a deck that provides physical proof of cheating? Half the players have side bets out, they don't double properly, and they are afraid to hit stiff totals. It is hard for me to watch, so I use the 'hide other players' option.

[Sidecount]

It would be helpful to know how the cards are checked before being put into play. How often are regulators watching this process in person? Do they record video every time and then subject to auditing? I have no idea how it works but I don't believe they are showing the cards face up at the table to prove the legitimacy of the decks like they typically do at a live casino.

[Archvaldor]

Hi OnlineAP,

The manipulation I'm talking about keeps the total number of cards in the shoe at 312 while altering the proportion of certain cards. For example, the house edge for a fair game of 6 decks, S17, DOA, DAS, SPA1, SPL3, NS is 0.4059% if players use a total-dependent basic strategy (TD-BS). In this scenario, by simply changing 2 cards in the shoe, the house edge increases to 0.6205%. If instead of 2, 4 cards are changed, the edge rises to 0.8365%, and if 6 cards are changed, the edge goes up to 1.0539%.

That’s why it’s very important to be able to analyze a quantity m of samples, each with a size n, that allows us to detect these changes and report these casinos if cheating is confirmed.

It’s also important for a player to know that a certain casino with live dealers is not doing anything dishonest.


Sincerely,
Cac

This issue is covered in Thorp's "Mathematics Of Gambling" pg 14 "Missing Cards: The Short Shoe". There are full text copies you can google online.

But this is all just basic statistical confidence test stuff, I'm pretty sure you know all this stuff, so maybe you are looking for something more elegant/efficient?

I'd add - and this sounds pedantic but I think it is important - it is impossible to know that a casino is not cheating. You can eliminate specific sources of cheating with statistical confidence but you can't eliminate all of them because they are theoretically infinite. You can only ever state casino cheating is unproven.