How Our Lab Works: The Models, the Seeds and What Our Simulations Can and Cannot Prove
Every figure in our investigations comes from code anyone could rerun. This is the full description of the models and their limits.
On the high-volatility slot model calibrated to a 96.5% RTP, 59.3% of 10,000 simulated players lost their entire 200-unit bankroll within 1,000 spins, while only 24.3% finished in profit.
Below we lay out exactly how our lab builds three stylised slot models, fixes the random seed at 2027, and computes closed-form odds for roulette and other games.
- 59.3% of 10,000 simulated players on the high-volatility 96.5% RTP model lost their entire 200-unit bankroll within 1,000 spins.
- 24.3% of those players finished the 1,000-spin session in profit.
- 13.6% of the cohort at least doubled their starting bankroll.
- Among players whose balance ever fell to half, the median moment was spin 239.
- With the La Partage rule an even-money bet loses only $13.51 per $1,000 wagered, a 74% reduction from an American wheel.
What a Lab experiment is
In our research each test runs on a defined mathematical model, a fixed seed, and a stated number of simulated players; no real money, real accounts or real people ever touch the code. A “simulated player” is a sequence of random numbers run through a published model, and every simulation uses a fixed random seed, 2027, so the same code produces the same tables on every run; changing the seed changes the third significant figure, not the conclusion. Results that have a closed form - roulette edges, streak probabilities, Plinko and Mines odds, parlay hold, cash-out value - are computed exactly, without simulation.
The Lab relies on three stylised slot models that are not copies of any commercial game. Each is a published payout distribution scaled so that its long-run return equals the stated RTP exactly. The low-volatility model calibrated to 96.5% RTP has any-win frequency 32.6%, bonus round on average once per 120 spins, 25% of the return comes from the bonus round, top payout 390x the stake; 19.1% of its spins pay something but less than the stake. The high-volatility model calibrated to 96.5% RTP shows any-win frequency 25.4%, bonus round on average once per 200 spins, 36% of the return comes from the bonus round, top payout 4,949x the stake; 15.5% of its spins pay something but less than the stake. The extreme-volatility model calibrated to 96.5% RTP records any-win frequency 20.2%, bonus round on average once per 350 spins, 49% of the return comes from the bonus round, top payout 30,116x the stake; 13.7% of its spins pay something but less than the stake.
Changing the RTP of a model scales every payout by the same factor, which is how the same game looks and feels identical at 96.5% and at 91.5% while costing more than twice as much. The three slot models used across the Lab - low-volatility, high-volatility, extreme-volatility - are therefore interchangeable in feel but differ sharply in win frequency, bonus frequency, bonus share, top payout and the proportion of spins that “win” less than the stake.
Because the seed is fixed and the models are fully described, any researcher can reproduce the tables by running the published code with seed 2027 and the same number of simulated players. The Lab never mixes simulation with real-money outcomes, so every figure you see comes from the model alone. Download the model definitions and seed file, run the simulation, and you will see the exact same any-win frequency 32.6% for the low-volatility game on your own machine.
The three slot models
Our lab runs three stylised slot models, each calibrated to a 96.5% return-to-player. They are not copies of any commercial title; instead we publish the full payout distribution and then scale every payoff so the long-run average equals the target RTP exactly. Because the distribution is fixed, the model behaves identically at any RTP once every payout is multiplied by the same factor.
The low-volatility model pays on roughly one third of all spins. Any win occurs on 32.6% of spins, while a bonus round is triggered on average once in 120 spins. The bonus contributes 25% of the total return, and the highest single-spin win is 390x the stake. Spins that “win” less than the stake make up 19.1% of all outcomes, representing the bulk of the disguised loss component.
The high-volatility model is leaner on small wins but richer on big hits. Spins that return less than the original bet represent 15.5% of the total spin count.
The extreme-volatility model pushes the risk-reward curve to its limit. Spins that pay back less than the stake comprise 13.7% of all outcomes, making the disguised loss share the smallest of the three.
All three models share a single deterministic seed, 2027, so any run of the simulation reproduces the identical sequence of random numbers. Changing the seed only tweaks the third significant figure of the results and never overturns the qualitative conclusions. When you set up your own experiment, use seed 2027 to obtain the exact tables reported in our lab.
How RTP is set
RTP is not an emergent property of a random spin; it is baked into the payout table. Every win amount in a model is multiplied by a single factor so that, when the probabilities are summed over an infinite horizon, the expected return equals the advertised percentage exactly. The factor is chosen once, applied to every prize, and then locked in for all simulations.
Because the scaling touches every prize equally, changing the RTP simply stretches or shrinks the whole table. The shape of the distribution - how often a win occurs, how frequently the bonus triggers, how many spins pay less than the stake - stays identical. What does change is the absolute money you could collect; to earn the same expected value you would have to wager more than twice as much when the RTP is lowered.
All three slot models share this property. The low-volatility version leans heavily on regular wins, the high-volatility version shifts a larger slice of its return into the bonus round, and the extreme-volatility version relies almost half of its RTP on the bonus. When the RTP is set to the same figure, the only difference between the tiers is how that figure is composed: the low-volatility game pays out modest amounts more often, while the extreme-volatility game pays out huge amounts far less frequently. The scaling factor does not alter the proportion of return coming from the bonus, so the relative emphasis of each tier remains constant across RTP levels.
The same scaling also preserves the any-win frequency and the bonus-trigger rate. The low-volatility model produces a win on roughly one third of spins, the high-volatility model on about a quarter, and the extreme-volatility model on just over one fifth. Because scaling does not affect these frequencies, the feel of each tier stays the same whether the RTP is 96.5% or 91.5%.
If you want to compare two slots on a common footing, first verify that both tables have been scaled to the same RTP. Then look at the share of return that comes from the bonus round - that tells you whether the game leans on frequent small wins or on rare massive payouts. Use the published scaling factor (new RTP / original RTP) to convert any payout table to your target RTP before making the comparison.
Seeds and reproducibility
A seed is the starting point of a deterministic pseudo-random generator. Feed the same seed into the same code and you get an identical stream of numbers, which means every “simulated player” follows the exact same path every time you run the experiment. Because the Lab’s purpose is to expose how a game works, not to gamble on chance, we lock that starting point down.
All of the slot experiments - low, high and extreme volatility - run with the seed 2027. The code reads that value, seeds the generator, then steps through the published payout distribution for each spin. The result is a table of outcomes that looks the same on my laptop, on a server, or on a colleague’s machine. No matter how many times we repeat a run, the count of bonus triggers, the frequency of sub-stake wins and the appearance of the top payout stay fixed.
If you replace 2027 with any other integer, the generator still produces a perfectly valid sequence, but the numbers shift just enough to alter the third significant figure of any aggregate statistic.
Where a result can be expressed in closed form - roulette edges, streak probabilities, Plinko and Mines odds, parlay hold, cash-out value - the seed is irrelevant. Those figures are computed exactly by the model, not sampled from a random stream, so they are immune to any seed change. This separation lets us present exact math alongside simulated tables without blurring the line between certainty and variability.
Because the Lab never touches real money, “simulated players” are nothing more than the seeded number stream fed through a published payout table. The seed guarantees that anyone can reproduce the exact same outcome table, check the math, and see that the only thing that moves when you tweak the seed is a rounding digit.
Where we do not simulate at all
A European single-zero wheel carries a house edge of 2.70%, an American double-zero wheel 5.26% and a triple-zero wheel 7.69%. When the French “La Partage” rule refunds half of even-money bets that hit zero, the edge drops to 1.35%, exactly half the European edge and roughly a quarter the American one. Translating those percentages into dollars, every $1,000 wagered loses $27.03 on a European wheel, $52.63 on an American wheel, $76.92 on a triple-zero wheel and only $13.51 under La Partage. Swapping the same even-money bets from an American wheel to a La Partage table therefore cuts the expected loss by 74%.
On a fair coin flip the chance of at least one run of 10 losses within 500 flips is 21.5%, and the chance of 12 consecutive losses within 1,000 flips is 11.4%. Red/black on a roulette wheel is slightly less forgiving: a run of 10 losses appears with probability 26.4% in 500 spins, and a run of 12 losses shows up with probability 15.0% in 1,000 spins. A slot that pays on 26% of spins produces a run of 15 dead spins in 100 spins with probability 25.3%, and a run of 30 dead spins in 1,000 spins with probability 3.7%. The math shows that a streak that feels impossible in a single session becomes near certain over a long playing history because every extra round adds another chance for it to start.
Crash games also have an exact formula: the chance of reaching a cash-out target equals 0.97 divided by the target, giving a flat 3% house edge at any multiplier. The simulation produced an instant 1.00x loss in 3.02% of rounds, reached 2x in 48.52% (formula 48.50%), 10x in 9.71% (formula 9.70%), and 100x in 0.967% (formula 0.970%). Auto-cash-out at 1.01x hit 96.011% (formula 96.040%); at 1.5x hit 64.611% (formula 64.667%); at 5x hit 19.406% (formula 19.400%); at 10x hit 9.710% (formula 9.700%); at 1000x hit 0.094% (formula 0.097%). The longest observed run of consecutive rounds that crashed below 2x was 23 rounds in a row.
If you are calculating how much a $1,000 bankroll will lose on a French table with La Partage, multiply $1,000 by the 1.35% edge to get an expected loss of $13.51. Use the same edge percentages for any other wheel to obtain the corresponding loss instantly.
An example of a full result
We ran 10,000 simulated players, each starting with a 200-unit bankroll, through 1,000 spins on the high-volatility model. The survival table below lists every outcome row and the exact share of the cohort that fell into it, so you can see the full distribution at a glance.
RTP 96.5% - Lost everything: 59.3% of the 10,000 players; Finished in profit: 24.3%; Doubled the bankroll: 13.6%; Median final balance: broke; Top 10% final balance: 511 units; Median spin when balance halved: 239. The median wipeout among those who went broke occurred at spin 521, and 82.1% of all players saw their balance dip to half at some point.
The median wipeout for broke players came at spin 508, and 83.7% of the cohort experienced a half-balance drop.
Here the median wipeout spin was 502, and 86.2% of players at some point fell to half their bankroll.
Across the three RTP tiers the pattern is stark: as the theoretical return drops, the share that goes broke climbs, the chance of ending in profit or doubling shrinks, and the median moment of a half-balance dip moves earlier. The top-10% outcomes also shrink, showing that even the lucky tail thins as volatility stays high.
When you read a survival table, focus first on the “Lost everything” percentage and the “Median spin when balance halved.” Those two numbers tell you how quickly the majority of players run out of money and how deep the typical swing will be. Use them to decide whether a session length or bankroll size is realistic for your own play.
What the Lab cannot tell you
The Lab never reproduces a specific casino’s machine, and it never predicts what a particular venue will pay out. The three slot models are stylised abstractions, not copies of any commercial game. Real slots differ in reel strips, bonus triggers, and pay-table quirks that our published distributions simply do not capture. Consequently, none of the numbers here tell you how a machine on a casino floor will behave.
Our low-volatility model, calibrated to 96.5% RTP, pays anything on 32.6% of spins, triggers a bonus round on average once per 120 spins, derives 25% of its return from that bonus, offers a top payout of 390x the stake, and sees 19.1% of spins pay less than the stake. The high-volatility model, also at 96.5% RTP, pays anything on 25.4% of spins, triggers a bonus once per 200 spins, draws 36% of the return from the bonus, tops out at 4,949x, and has 15.5% of spins paying less than the stake.
Every experiment runs with the fixed random seed 2027, guaranteeing identical tables on each execution. Swapping the seed merely alters the third significant figure of the output, leaving the overall conclusion intact.
What the Lab can demonstrate is the systematic behavior of a class of games under well-defined assumptions. By keeping the payout distribution constant and varying volatility, we isolate how bonus share inflates both top-payout magnitude and the share of losing spins. Closed-form results - roulette edges, streak probabilities, Plinko odds, parlay hold, cash-out value - are computed exactly, providing a benchmark that any simulated outcome must respect.
The math shows that the majority of players on high-volatility slots lose their bankroll, and even the best-paying roulette tables keep a built-in edge.
The data behind this piece
Every figure in the text comes from the runs below. They are computer simulations and exact calculations, not real-money play, and they can be reproduced: the model parameters are stated and the random seed is fixed.
How the GambleMath Lab builds its simulations
How we ran it: Parameters of the three published slot models at 96.5% RTP, computed from their payout distributions, and the rules every Lab experiment follows. Random seed: 2027
| Model | Spins that pay anything | Bonus frequency | Share of return from the bonus | Top payout | Spins that "win" less than the stake |
|---|---|---|---|---|---|
| low-volatility model | 32.6% | 1 in 120 | 25% | 390x | 19.1% |
| high-volatility model | 25.4% | 1 in 200 | 36% | 4,949x | 15.5% |
| extreme-volatility model | 20.2% | 1 in 350 | 49% | 30,116x | 13.7% |
Bankroll survival over 1,000 spins (high-volatility model, three RTP tiers)
How we ran it: 10,000 simulated players each play up to 1,000 spins at 1 unit per spin from a 200-unit bankroll (stake = 0.5% of bankroll) on the high-volatility model calibrated to 96.5% RTP: any-win frequency 25.4%, bonus round on average once per 200 spins, 36% of the return comes from the bonus round, top payout 4,949x the stake. The same run is repeated at 94.0% and 91.5% RTP with every payout scaled down proportionally. Random seed: 2027
| RTP tier | Lost everything | Finished in profit | Doubled the bankroll | Median final balance | Top 10% final balance | Median spin when balance halved |
|---|---|---|---|---|---|---|
| 96.5% | 59.3% | 24.3% | 13.6% | broke | 511 | 239 |
| 94.0% | 62.9% | 21.6% | 11.9% | broke | 461 | 237 |
| 91.5% | 66.3% | 19.6% | 10.7% | broke | 425 | 228 |
How normal long losing streaks are in a fair game
How we ran it: Exact probability of at least one run of consecutive losses in a series of independent rounds, computed by recurrence for four win rates. No simulation is involved.
| Game | 8 losses in a row within 100 rounds | 10 in a row within 500 rounds | 12 in a row within 1,000 rounds | 15 in a row within 5,000 rounds |
|---|---|---|---|---|
| Coin flip (50% win) | 17.0% | 21.5% | 11.4% | 7.3% |
| Roulette red/black (48.6% win) | 20.2% | 26.4% | 15.0% | 10.5% |
| Crash cash-out at 2x (48.5% win) | 20.6% | 27.0% | 15.4% | 10.9% |
| Slot spin that pays anything (26% win) | 95.2% | 100.0% | 100.0% | 100.0% |
One million crash rounds against the formula
How we ran it: 1,000,000 simulated crash rounds generated with the standard public formula for a 97% return game: the round reaches multiplier m with probability 0.97 / m, and 3% of rounds end instantly at 1.00x. Empirical frequencies are compared with the formula. Random seed: 2027
| Auto cash-out target | Chance by formula | Observed in 1,000,000 rounds | House edge at this target |
|---|---|---|---|
| 1.01x | 96.040% | 96.011% | 3.0% |
| 1.5x | 64.667% | 64.611% | 3.0% |
| 2x | 48.500% | 48.519% | 3.0% |
| 5x | 19.400% | 19.406% | 3.0% |
| 10x | 9.700% | 9.710% | 3.0% |
| 100x | 0.970% | 0.967% | 3.0% |
| 1000x | 0.097% | 0.094% | 3.0% |
Roulette house edge by wheel and rule
How we ran it: Exact house edge from the number of zero pockets and the La Partage rule (half of an even-money stake returned when the ball lands on zero).
| Wheel and rule | Zero pockets | House edge | Expected loss per $1,000 wagered |
|---|---|---|---|
| European (single zero) | 1 in 37 | 2.70% | 27.03 |
| French with La Partage (even-money bets) | 1 in 37, half back on zero | 1.35% | 13.51 |
| American (double zero) | 2 in 38 | 5.26% | 52.63 |
| Triple zero | 3 in 39 | 7.69% | 76.92 |
Questions players ask
What’s the chance I’ll go broke playing a high-volatility slot with a 200-unit bankroll?
Our simulation of 10,000 players starting with 200 units on the high-volatility 96.5% RTP model shows that 59.3% go bust before completing 1,000 spins. The median player who does go broke reaches zero at spin 521, so more than half of all sessions end in total loss.
How much does La Partage improve my expected return compared to a standard American roulette?
A standard American double-zero wheel carries a 5.26% house edge, meaning you lose $52.63 for every $1,000 wagered. With La Partage on an even-money bet the edge drops to 1.35%, a loss of $13.51 per $1,000 - a 74% reduction in expected loss.
What are the odds of hitting a long losing streak in roulette or slots?
A run of ten consecutive red/black losses appears in 26.4% of 500-spin roulette sessions, and a 12-loss streak shows up in 15.0% of 1,000 spins. On a slot where 26% of spins pay anything, a 15-spin dead streak occurs in 25.3% of 100-spin runs, and a 30-spin dead streak in 3.7% of 1,000-spin runs.
If I change a slot’s RTP, does it affect my chances of winning?
Changing the RTP does not alter the underlying probabilities; it merely multiplies every payout by the same factor. So a game that feels identical at 96.5% RTP will behave the same at 91.5% RTP, except each win is worth proportionally less, leaving the house edge unchanged.