What one million simulated roulette sessions taught us
The strongest findings from simulating ten betting systems for one hundred thousand sessions each: every system converged to the house edge, a high win rate said nothing about the bad days, and streaks run long.
Published 2026-09-23, from the dataset generated the same day.
The setup
We took the ten classic betting systems shipped in the app and ran each through 100,000 simulated sessions of up to 100 spins on a fair single zero wheel: one million sessions, all dealt the same seeded spins, all played by the exact strategy code the app runs. Every figure below comes from that one dataset.
Finding one: nobody escapes the edge, and the misses are tiny
The edge predicts each system's loss from one number: what it staked, divided by 37. Here is prediction against reality, per session, and the ratio between them.
| System | Avg staked | Edge predicts | Simulation got | Ratio |
|---|---|---|---|---|
| Flat betting | 500 points | -13.5 points | -13.6 points | 1.00 |
| Martingale | 1,159.7 points | -31.3 points | -31.1 points | 0.99 |
| Grand Martingale | 1,419.9 points | -38.4 points | -38.4 points | 1.00 |
| Paroli | 843.6 points | -22.8 points | -22.9 points | 1.00 |
| D'Alembert | 2,318.1 points | -62.7 points | -63.4 points | 1.01 |
| Reverse D'Alembert | 2,594 points | -70.1 points | -70 points | 1.00 |
| Fibonacci | 1,071.6 points | -29 points | -29.3 points | 1.01 |
| Labouchere | 3,159.5 points | -85.4 points | -84.2 points | 0.99 |
| Oscar's Grind | 1,291.8 points | -34.9 points | -35.4 points | 1.01 |
| James Bond | 6,037.3 points | -163.2 points | -162.8 points | 1.00 |
Ten systems, wildly different personalities, every ratio hugging 1.00. The systems chose how the losing was distributed; the wheel chose how much there was. This one table is the whole argument against betting systems, told by the systems themselves.
Download the chart as a PNG, 2400 by 1350 pixels or 1200 by 675 pixels, or the plotted figures as a CSV file.
Finding two: a win rate says nothing about the bad day
The number a system is sold on, its session win rate, told us nothing about how bad its bad sessions were. Fibonacci finished ahead in 73.8% of sessions, and still had a worst session of -440 points and a 23.0% bust rate. Flat betting won only 35.6% of sessions, and its worst case was just -210 points. The Martingale sat in the middle of the win column at 49.2% while busting 50.1% of the time. Win rate is not a quality score; it is a style score, and the best strategy page ranks all ten through every lens. Our own original system obeys the same law, published on its page with the same honesty.
Finding three: streaks run longer than intuition
Betting red every spin, the average session's longest losing run was 6.2 spins, and the worst single session endured 24 in a row. Runs like that are why doubling systems die: they are rare per session and routine across a hundred thousand of them. The gambler's fallacy page has the exact streak math.
What we did with these findings
We published everything: the distributions, the busts, the worst sessions, the average losses, for every system including our own. Not because the data found an exploit, it found the opposite, but because the absence of an exploit, shown plainly, is the most useful thing a roulette site can say.
Check the numbers
Every underlying number is on the strategy pages, and the app replays any system on the live table, so you can watch the same edge at work.