Gamdom Crash audit: house edge up to 2.98%, not the advertised 1%

[Alex077]

On November 1, independent auditor ProvablyFair.org published an investigation into Gamdom's in-house Crash game. Gamdom advertises 99% RTP (1% edge)...

Frankly this explanation needs to be dumbed down a level further or better explained. I follow the general gist and supposedly "provably fair" games have been discovered as wrong in the past, but certain parts of the explanation - like how you split out the multipliers are hard to follow by somebody who doesn't do that much gambling on crash games. It seems like if they're going to go to all this trouble of rigging the game in a certain way they have screwed up royally by publishing all the outputs on the front page and not throwing in some basic fake bets to cover the scam up, but it's very possible...

In my opinion, we need to separate two issues here: the possible difference between the actual RTP and the advertised 99%, and the lack of transparency in the algorithm that determines the game results.

At the same time, it's important to understand that even a small difference in RTP can make a huge difference when we're talking about large betting volumes. For example, with a total wagering volume of $100 million, a 1% house edge would theoretically bring the casino $1 million, while a 3% house edge would bring in $3 million. Otherwise, the whole Provably Fair label pretty much loses its meaning.

[leea-1334]

I understand the basic math,,, I wish I understood the exact way it was explained, but I think there is a minimum conclusion here that 1% House Edge is extremely misleading, I really hate when the house tries to use rakeback to form part of the HE as this in no way is accurate.

To be fair, I think very many sites don't reach their advertised house edge of whatever %. Spoke about that before. I have 10.5 million wager on stake for example, 9 million of that comes from the house games, round about. A while ago I checked my personal RTP for every house game I ever played there and found that not in 1 single game I have reached the advertised RTP, not in 1. Even in the one with the most bets/wager (round about 400000 bets) my RTP is still 0.7% under 99%. I mean 400k bets are already a pretty healthy sample size and 0.7% doesn't sound like much, but it is. With about 5 million wager on that game a 0.7% deviation means 35000$ in more losses, that's huge in the end.

Yes I think I found the same in almost every game I checked in the past on Stake originals,,, however, HE might not translate directly to RTP as we expect, even with 400k bets,,, in almost every dice site I used I had over 10 million and that was much closer reflection than 1 million. And player RTPs is never a good way to show PF but for sure,,, I do not trust Stake does it well, but Gamdom for sure clearly is doing it wrong

[swogerino]

Imagine now me that have been opening quite a few threads not for a casino specifically but for a specific provider being Pragmatic Play who I think pays money to auditors as there is no other way explaining 29 buy bonuses in a row resulting in being losing ones. Sure the story here maybe true and I think auditors are not doing a good enough job for many of slot providers or crash originals so these things it is only natural for such complaints to be happening more often nowadays.

[Betwrong]

To be fair, I think very many sites don't reach their advertised house edge of whatever %. Spoke about that before. I have 10.5 million wager on stake for example, 9 million of that comes from the house games, round about. A while ago I checked my personal RTP for every house game I ever played there and found that not in 1 single game I have reached the advertised RTP, not in 1. Even in the one with the most bets/wager (round about 400000 bets) my RTP is still 0.7% under 99%. I mean 400k bets are already a pretty healthy sample size and 0.7% doesn't sound like much, but it is. With about 5 million wager on that game a 0.7% deviation means 35000$ in more losses, that's huge in the end.

That's actually a good result. Thanks for your research! I personally think that with 4 million bets your personal RTP would become closer to the advertised RTP - the difference would be around 0.1%, and with 40 million bets it would be almost zero, but my question is, do you really care about 0.7%? I mean, how does it affect you? Are you making money from playing house games and see that you are 0.7% under or what? Anyways, it's not for the first time I'm reading about your gambling and I 'm thankful for your reports. Thanks, man! I really value them.

[DICE777]

my question is, do you really care about 0.7%? I mean, how does it affect you? Are you making money from playing house games and see that you are 0.7% under or what?

Fair question for a personal RTP graph, but I think that 0.7% gap is a different problem from this thread.

If you play hundreds of thousands of bets on Stake originals and land 0.7% under the advertised RTP, that can still be variance. Like you said, more volume should pull the number closer. Player RTP is a noisy way to judge a game.

Gamdom Crash is cleaner than that, because you don't need a million-bet sample. The extra edge is sitting in the published verifier formula before anyone plays: a 2x target pays 98%, not 99%. That's an extra $10 expected cost per $1,000 wagered at 2x, and closer to $18 per $1,000 if you're hunting 10x. When Gambafish recomputed 6.8M+ rounds of the current chain and got ~98.0% at 2x, that wasn't an unlucky sample. That's the formula showing up in the results.

I think there is a minimum conclusion here that 1% House Edge is extremely misleading, I really hate when the house tries to use rakeback to form part of the HE as this in no way is accurate.

Yeah, this is the part that still bothers me. The Zero Edge rakeback came back at exactly 1% per bet on the account that was tested, so a 2x player gets lifted to 99%, not to the 100% the promo implies, and a 10x player only gets to about 98.21%. Romer saying Originals ran above 100% RTP this year overall can still be true if rakeback, other games and a few big sessions pad the blend. It just doesn't answer the formula question.

So I do care about the gap when it's baked into the code and the help text still says 1%. Sample-size arguments make sense for someone's personal graph. They don't rewrite a published RTP(x).

Has anyone here who actually plays Gamdom Crash checked their own cash-out mix against that table? Curious if low-target players notice it less, which is what the maths predicts.

[General_Bitcoin]

It would be much more time saving if casinos opted to develop free source originals, there would be no need for someone to run a time and resource consuming audit just to get an approximation on true values.

Yes, open-code would help with the transparency, but there's more to it than that. If players know the code is open-source they still need to know that it is being run by the casino on their servers. What would really help is the entire system being independently verifiable. This means players can view the formula, and how results are obtained. At the moment casinos can publish clear code for everyone to see, but players have no easy way of knowing what is going on.

[DICE777]

Yes, open-code would help with the transparency, but there's more to it than that. If players know the code is open-source they still need to know that it is being run by the casino on their servers. What would really help is the entire system being independently verifiable.

Agree, and this case is a pretty clean example of why those are two different checks.

Published code was enough for the RTP part. Gamdom already showed a verifier, and that's how the extra edge showed up: the formula itself pays 98% at 2x, not the advertised 99%. You don't need to trust their servers for that finding. Anyone can plug a target into

Code:

RTP(x) = 0.99 * 0.98 * x / (x - 0.02)

and see the gap.

The second claim in the report is closer to what you're describing. Every replayed round matched the published code, so the results are reproducible. That still doesn't prove the sequence was chosen fairly. Crash moved to a new hash chain sometime between Oct 9 and Oct 26 with no new public commitment found, and both chains use the hash of Bitcoin block 339,300 (from 2015) as a salt. Because that salt was already public, an operator could in theory pre-compute many candidate chains and pick one. Each result would still pass the verifier.

That's the "are they running what they published, and did they commit to it before play?" problem. The usual fix is: publish the commitment first, then salt with a future Bitcoin block mined after that commitment. After that, a player can verify a round from public data without taking the server's word for it.

So I wouldn't treat open-source originals as a replacement for audits, and I wouldn't treat a public verifier as the whole PF package either. One check answers "does this formula match the advertised 1%?" The other answers "can I prove the chain wasn't picked after the fact?" This report failed the first one in the maths and the second one on missing proof. It still didn't show the chain was manipulated.

Would you count a game as independently verifiable only if it uses that commit-then-future-block setup, or is a public verifier plus replay enough for you as long as the RTP formula is honest?

[MaxRoman]

Would you count a game as independently verifiable only if it uses that commit-then-future-block setup, or is a public verifier plus replay enough for you as long as the RTP formula is honest?

For me a public verifier plus replay only covers half of it. It shows the maths is honest and that the results match the published chain but it can't show the chain was locked in before anyone played. So I'd only call a game independently verifiable if the commitment goes out first and the salt comes from a Bitcoin block mined after it, otherwise pre-computing candidate chains stays possible even with an honest RTP formula.

[theoddwright]

Even in the one with the most bets/wager (round about 400000 bets) my RTP is still 0.7% under 99%. I mean 400k bets are already a pretty healthy sample size

Whether 400k bets is a big sample depends on what you were betting on. The noise in your RTP is roughly the per-bet spread divided by sqrt(400,000) ≈ 632, and the per-bet spread grows with the multiplier.

For equal bets with a 1% edge, one standard deviation of your RTP after 400k bets is about:
2x target: 0.16%
10x: 0.47%
100x: 1.6%

So if that game was mostly 2x dice, being 0.7% under is more than 4 standard deviations, and I'd want to look at it. At 10x it's 1.5 SD, which happens to about one player in fifteen. At 100x it's nothing at all.

Two things make it wider still. RTP is weighted by wager, so if your bet size changed, a few big bets weigh as much as thousands of small ones. And the noise only shrinks with the square root: ten times the bets makes it about three times smaller, not ten.

[AHOYBRAUSE]

Even in the one with the most bets/wager (round about 400000 bets) my RTP is still 0.7% under 99%. I mean 400k bets are already a pretty healthy sample size

Whether 400k bets is a big sample depends on what you were betting on. The noise in your RTP is roughly the per-bet spread divided by sqrt(400,000) ≈ 632, and the per-bet spread grows with the multiplier.

For equal bets with a 1% edge, one standard deviation of your RTP after 400k bets is about:
2x target: 0.16%
10x: 0.47%
100x: 1.6%

So if that game was mostly 2x dice, being 0.7% under is more than 4 standard deviations, and I'd want to look at it. At 10x it's 1.5 SD, which happens to about one player in fifteen. At 100x it's nothing at all.

Two things make it wider still. RTP is weighted by wager, so if your bet size changed, a few big bets weigh as much as thousands of small ones. And the noise only shrinks with the square root: ten times the bets makes it about three times smaller, not ten.

Thanks for the explanation. Yeah basically dice wise I always bet 2x dice, or even 1.90 dice to be precise, around 52% win chance bets always. Never anything else on this game. So I guess my bad feeling really is warranted that the deviation seems a bit blown up. But yes, obviously the bet size changed from time to time, being all over the place.