1/ We investigated @bcgame's Video Poker
We found a biased shuffle that affects every hand dealt and lowers players expected returns.
Advertised: 2% house edge
Actual: ≈3.6%
Around 80% more expected player losses than advertised, under optimal fair deck play.
The card order responsible appears in BC’s published code from April 2020. Anyone who played Video Poker since then may have been affected.
We’re asking BC Game to address this publicly and compensate affected players.
Full breakdown below.
The CEO of @bcgame says he’ll take a look
80% higher house edge than advertised on BC Game Video Poker, from a shuffle bias in its published code for 6+ years
Will BC Game fix it and compensate players for the full period? We’ll track the response
1/ We investigated @bcgame's Video Poker
We found a biased shuffle that affects every hand dealt and lowers players expected returns.
Advertised: 2% house edge
Actual: ≈3.6%
Around 80% more expected player losses than advertised, under optimal fair deck play.
The card order responsible appears in BC’s published code from April 2020. Anyone who played Video Poker since then may have been affected.
We’re asking BC Game to address this publicly and compensate affected players.
Full breakdown below.
7/ We’re asking @bcgame to:
-Respond publicly and address the shuffle and return discrepancies.
-Replace the biased shuffle and publish an updated verifier.
-Publish when each version went live and which paytables applied.
-Compensate affected players using its complete betting records.
If you’ve played Video Poker since April 2020, or the Player Pair bet in its single player Baccarat, keep your bet history and IDs and ask BC Game to review your wagering.
6/ We also found a problem with the Player Pair bet in BC’s single player Baccarat, which uses the same rotate and sort approach.
At the same payouts:
Fairly shuffled two-deck shoe: 80.81% return
BC’s dealing method: about 77.14%
That means about $37 more expected loss per $1,000 wagered on Player Pair.
We tested 150 live Baccarat rounds. Every round matched BC’s published verifier. The return calculation uses 40 million simulated rounds.
The largest shortfall is on Player Pair. The other bets changed by much smaller amounts, in different directions.
Banker Pair moves the other way, improving by about 0.57 percentage points. That partly offsets the Player Pair shortfall for players who bet both.
5/ A 2020 change made the odds worse
BC’s original dealing method favored players in our simulations: about 102.6% return.
An update to its published code dated 8 April 2020 changed the card list. With the same paytable and playing strategy, the return dropped to about 97.3%.
That replacement list is still used today.
Using this list, all three paytables we documented return less than advertised:
99% version (documented from 2020 to early 2025): Advertised: 1% house edge
Actual: ≈2.7%
96% version (documented in 2025):
Advertised: 4% house edge
Actual: ≈5.5%
98% version (current):
Advertised: 2% house edge
Actual: ≈ 3.6%
All with optimal fair deck play.
The code history and historical player bets support a longstanding issue. These dates come from public records; exact switch dates need BC’s records.
4/ The verifier passes. The shuffle is still biased.
We played 62 real money hands, using every possible hold at least once, from keeping nothing to keeping all 5.
Every hand matched BC’s published verifier, including all 10 cards in order. The revealed server seeds hashed to the commitments recorded before play.
That confirms the sampled games follow the published dealing method.
It doesn’t establish that the method shuffles fairly.
To calculate the return, we worked out the best fair deck hold for every possible starting hand, then used that strategy across 600 million simulated BC deals.
Result: 96.40%, with a 95% interval of 96.37–96.44%.
A fair shuffle control matched the expected return.
3/ What it costs players
For every $10,000 wagered on the current game:
Advertised expected loss: $200
Expected loss using BC’s deal: about $360
Extra expected loss: about $160
That’s using the best holding decisions for a fairly shuffled deck and the game’s paytable.
These are long run averages. Wagering means the total of your bets, including winnings bet again.
The extra loss comes from the dealing method. With the same paytable and holding strategy, a fair shuffle returns 98.02%. BC’s deal returns about 96.4%.
2/ What’s wrong with the shuffle?
BC sorts its cards using shifted copies of the same random string. This links their positions.
Cards next to each other in BC’s current internal list appear together in the opening hand about 30% more often than with a fair shuffle. But those neighbors never share a rank or suit.
That makes valuable starting hands less common.
Across 600 million simulated deals, these starting hands appeared less often than with a fair shuffle:
Royal flush: 19.1% fewer
Straight flush: 13.6% fewer
Flush: 13.0% fewer
Four of a kind: 7.4% fewer
Full house: 3.7% fewer
Three of a kind: 3.6% fewer
These figures are for the first five cards, before any hold or replacement. Pairs and flushes can still happen. Their probabilities have changed.
All 6 in-house games @LIQDcom submitted for audit are now certified by @provablyfairorg
We placed 44,134 bets anonymously, simulated 210M rounds, and ran 201 game integrity tests.
We tested each game’s RTP, odds, RNG, and provably fair implementation. Our independent rebuild reproduced every audited bet exactly.
All 6 games worked as advertised in our tests.
Our tests are open source. Anyone can check every audit bet and finding.
Full audit and test repos below.
A lot of people have asked about the motives behind our game investigations. Is someone paying us to target casinos? How do we make money?
Nobody commissions or pays us to produce an investigation and we don’t accept payment to target a casino. Most investigations start with a player messaging us about something they noticed in a game
We fund the work ourselves, spend dozens of hours on each one and only publish a finding when we have reproducible evidence to back it up.
We’ve found issues at several casinos. We explain what happened without guessing the casino’s intent.
In the past week I’ve had 100s of messages from players asking whether specific games they play actually work as advertised. We can’t check every game at every casino, but we’ll keep investigating and sharing what we find publicly.
Publishing our findings means players get to see what’s happening in games they’re putting real money into. It also shows casinos why it’s worth having someone independent check their games, which is a commercial benefit to us.
How do we make money? Casinos can hire us to audit their in house provably fair games. An audit can catch a lot, from a game giving players worse odds than advertised to a bug handing players an edge the casino never intended.
We charge a flat fee for audits and certification, set before any testing starts and the same whether the games pass or fail. We don’t take affiliate commissions, revenue share or referral fees from casinos, so we earn nothing from how much players wager or lose.
We publish what we find, including issues fixed during an audit. A game that isn’t fair can’t pay its way to a pass. Our audits are runnable code and open source, so anyone can reproduce or check the results.
The problem we’re trying to highlight is that casinos build their own games and usually nobody independently checks them. Issues that cost players can go unnoticed for years.
Credit to @Gamdom for fixing the Crash formula. They used the correction we proposed and every target now gives a 99% expected return.
But their compensation only checks your saved target, even if you usually cash out earlier. And Crash still isn’t provably fair as implemented.
HOW GAMDOM WORKED OUT COMPENSATION
Gamdom says it replayed old rounds with the new formula. For each bet, it checked one thing: would the saved auto cash out target have been reached? If yes, the bet was paid as a win.
Say your target was 2x. About 1 round in 200 crashed at 1.98x or 1.99x. Under the new formula those rounds reach 2.00x to 2.02x, so a bet waiting for 2x now gets paid $2 per $1, stake included. For players who let their target ride, that’s fair. It works out to about $10 per $1,000 wagered at 2x, which is what the old odds cost them.
WHY MANY PLAYERS GOT NOTHING
Crash lets you cash out manually at any point before your target and many players do.
Say you kept 2x saved but usually clicked out at 1.5x. About 1 round in 225 crashed at 1.49x and would now reach 1.50x or 1.51x, giving you the chance to click.
The replay only checks your auto 2x target, so you get nothing for those rounds. Yet you paid the higher house edge on every bet.
In our 44 hour recording, the 7 biggest players accounted for 46% of all wagering. Each cashed out at more than 80 different multipliers. Only Gamdom’s records can show how often they cashed out below their saved target.
WHY OUR NUMBER IS HIGHER
We priced the extra edge on every bet, however it was cashed out, because every bet faced the worse odds: $6.69 per $1,000 at 1.5x, $10 at 2x, $17.86 at 10x.
At the pace we recorded, that’s roughly 16.5 million bets since January and an estimated $1.58M in extra expected losses. Gamdom reported $339k in balance adjustments across 1,293 accounts.
The number of bets is estimated from a 44 hour window, so the real total could be lower or higher.
How to compensate is Gamdom’s call. We’re just explaining where the difference comes from.
PROVABLY FAIR ISSUE IS STILL OPEN
Gamdom switched Crash to a new hash chain twice: on 15 October, and again when the formula fix went live on 3 November. We found no public notice of either chain change and no matching advance commitment for either new chain.
The game still uses a Bitcoin block hash known since 2015. With that input already public, an operator could theoretically compare chains in advance and choose one that favors the house, and those results would still verify. We haven’t established that Gamdom did this.
But for the game to be provably fair, it needs fixing too: publish the commitment before a new chain starts, name a future Bitcoin block in advance, and use its hash as the salt, not one from 2015. If Gamdom does this, we’ll verify it and say so publicly.
As @provablyfairorg reported, the actual house edge on our Crash game was higher than the advertised 1%, ranging between 1% and 3% depending on the player's target multiplier.
We've now changed the formula so it applies a flat 1% house edge across all target multipliers. The instant-crash rate remains unchanged, while all other crash points scale up slightly. The new formula can only produce an equal or higher result than the previous formula, meaning every historical round continues to verify against its original seeds.
We re-ran historical rounds using this new formula and wherever an auto-cashout target would now have been reached, we've recorded the bet as a win and credited the corresponding balance.
Total adjustment: $339,113.41 across 1,293 accounts. On top of this, we've allocated a further $250,000 in reload credits to those same 1,293 accounts, distributed proportionally to their individual adjustments.
7/ We’re asking @Gamdom to:
- Respond publicly and explain why Crash’s odds don’t match the advertised house edge.
- Make Crash return the advertised 99% on average at every cash-out target, or clearly advertise the real edge.
- Refund the extra edge on affected bets.
- Fix the provably fair setup.