Online crypto casinos publish return-to-player (RTP) figures for their in-house blackjack games that sit close to the theoretical ceiling for the game – mostly between 98.4% and 100%. But data from the on-chain gambling tracker gamstat.io, drawn from roughly 2.7 million captured blackjack rounds across 15 major casinos, shows a persistent gap between those declared figures and what the platform actually measured from live betting feeds. In several cases the gap is enormous: tens of percentage points, not fractions of one.
Numbers
Across the 15 casinos gamstat tracked, declared blackjack RTP clusters tightly (98.4%–100%). Observed RTP is scattered and, with one exception, consistently lower:
| Casino | Declared | Observed |
|---|---|---|
| Stake | 99.43% | 81.22% |
| Duel | 100.00% | 89.21% |
| Roobet | 99.50% | 97.77% |
| Shuffle | 99.48% | 76.89% |
| Gamdom | 99.31% | 84.37% |
| Rainbet | 99.60% | 101.02% |
| Thrill | 99.48% | 93.23% |
| Duelbits | 99.29% | 49.03% ⚠ |
| Winna | 98.41% | 54.16% ⚠ |
| Yeet | 99.60% | 83.17% |
| BitStarz | 99.50% | 40.81% ⚠ |
| BC.Game | 99.52% | 78.28% |
| Cloudbet | 99.48% | 95.30% |
⚠ = gamstat flags these as low-confidence
Only Rainbet’s observed figure exceeds its declared one. This dataset, and the framing that “not a single group of players is in profit overall” except on one site, is what circulated widely on X this week, sourced from a post by user @xet linking directly to gamstat’s live tracker.
Explanation 1: Players Don’t Play Optimally
Declared RTP for blackjack is a theoretical figure calculated under perfect basic strategy – the mathematically correct play on every single hand, every split, every double, every stand. Almost no population of real players, anywhere, achieves that. Real players stand on stiff hands out of instinct, split tens, double on weak totals, and misplay soft hands. This is a well-established and substantial source of return leakage in blackjack, and it applies equally to a physical casino floor and an online one – it is not specific to crypto casinos or in-house software.
This is the single most common explanation offered by critics of the “rigged” framing, and it’s likely a real contributor. What it can’t easily explain on its own is the size of some of these gaps. Player misplay in blackjack typically costs a percentage point or two off the optimal edge, not the 10–20+ percentage point swings seen at Stake, Shuffle, Duelbits, or BitStarz. Misplay alone is a plausible partial explanation; it’s a strained explanation for the largest gaps in this table.
Among them, there are surely some idiots who split tens, who stay on 15 and double down on 12; it’s a crappy study you’ve put out. If you do this with people who do the blackjack theory at 100%, the RTP will be good that’s a fact.
— TeufeurS (@teufeurs) August 8, 2026
Explanation 2: The Sample Itself Is Small
The sample is smaller than it sounds: 2.7 million rounds across 15 casinos works out to about 180,000 hands per casino, which isn’t huge for a game with such a thin house edge. Even @xet admitted as much in the replies.
The bigger issue is how the data is collected. Gamstat scrapes each casino’s public “live feed,” a stream of recent bets. So this looks more like “whatever the scraper happened to see” than an audited return figure.
As someone who works with casino games. Ur sample size needs to be way larger to get anything accurate. It’s weird that none of them are above 100% because they should given the variance. But that could also just be variance or human error.
— Variance Bro (@VarianceBro) August 9, 2026
Explanation 3: The Games Are Returning Less than Advertised
This is the explanation implied by the “rigged” framing. Most of these games claim to be “provably fair,” meaning the RNG seed and shuffle can in principle be independently verified round-by-round against the casino’s committed hash. That verification – actually replaying seeds against dealt cards to check for tampering – is the kind of evidence that would distinguish “the game is unfair” from “the sample is small and skewed.” No such verification has been published alongside this data.
I gambled online probably 1000 hours plus… its insane how rigged online casinos are…. Go in reallife in casino its way better for you bankroll and your chances are higher also
— BrainlessApe (@BrainlessApe420) August 9, 2026
Where This Leaves the “Rigged” Question
The gap between declared and observed RTP in this dataset is real and, at the extremes, striking – a 49% or 41% observed return against a ~99% declared figure is not a rounding error under any of the three explanations above. But the size and consistency of the effect look more like what you’d expect from a feed that skews toward certain rounds (explanation 2), compounded by ordinary player misplay (explanation 1), than from uniform underpayment across 15 independently-run games (explanation 3) – precisely because gamstat’s own documentation predicts a feed-driven distortion of this kind and flags several of the most extreme readings (Duelbits, Winna, BitStarz) as low-confidence.
The fair summary: this data shows a real and worth-investigating discrepancy between what these games promise and what a partial, admittedly-imperfect tracking effort has captured. It does not, on its own, establish that any specific casino’s blackjack implementation is rigged. Closing that gap would require either a much larger, wager-weighted sample, or independent seed verification against actual dealt hands – neither of which has been done publicly yet.
