Original research, September 2026
A million roulette spins, and what they really show
We ran one million roulette spins, 20,000 betting-system sessions and 300,000 slot sessions to test the claims players argue about most: hot numbers, overdue colours, the Martingale, and how often a slot session really ends in profit.
48.62%
Red came up after five reds in a row. Its normal rate is 48.65%.
24%
of Martingale sessions lost the whole £200 budget.
21%
of high-volatility slot sessions finished ahead, against 39% on low volatility.
19
reds or blacks in a row, the longest streak in a million spins.
What did we simulate, and how?
Let's start with the method, because research is only as good as the way it was done. Everything on this page was generated with Python and NumPy using a fixed random seed, 20260927, so anyone can run the same code and get exactly the same numbers. There are three experiments.
- Roulette: 1,000,000 spins of a single-zero European wheel with 37 pockets.
- Betting systems: 10,000 Martingale sessions and 10,000 flat-betting sessions on red, each starting with £200 and a £1 base stake, over at most 500 spins. Martingale sessions stopped on reaching £50 profit or running out of money, with a £500 table maximum.
- Slots: 100,000 sessions of 200 spins at £1 on each of three model slots, all paying back exactly 96% but with low, medium and high volatility.
The model slots aren't copies of real games. They're simple pay tables built to show one thing clearly, which is how volatility changes a session when the RTP stays the same. Their full pay tables are in the downloadable code.
Does a fair roulette wheel even out over a million spins?
Yes, closely, but never perfectly. Each number should land about 27,027 times in a million spins. The least frequent, number 4, landed 26,747 times, and the most frequent, number 6, landed 27,420 times. Every number was within about 1.5% of its expected count.
That spread is exactly what randomness looks like. A chi-square test, the standard way to check whether counts like these are consistent with a fair wheel, gave a value of 28.27 on 36 degrees of freedom. That's comfortably normal, with a p-value of about 0.82. In plain English: if you'd tracked these numbers at a real table and called number 6 "hot", you'd have been reading meaning into noise. The bias wheel checker runs the same test on your own spin data.
Is red due after a run of blacks, or reds?
No, and a million spins makes the point very clearly. We looked at every moment when red had just come up five times in a row. It happened 27,208 times. On the very next spin, red came up 48.62% of the time. Red's normal rate on a European wheel, 18 pockets out of 37, is 48.65%.
The difference is a rounding error. The wheel has no memory, so a streak tells you nothing about the next spin. This is the gambler's fallacy, and it's the idea behind most "tracker" and "predictor" apps sold to roulette players. The roulette AI page covers why no app can get round it.
How long can streaks and droughts get?
Longer than most people expect. The longest run of the same colour in our million spins was 19 in a row. The longest time a single number went without landing was 520 spins, for number 16, against an average gap of 36.
These aren't signs of anything wrong. Over a long enough period, rare streaks are guaranteed to happen somewhere. The danger is betting more because a streak "has to end". That's exactly what the next experiment tests.
Does the Martingale system work?
It works most of the time and fails badly the rest of the time, which averages out to a loss. The Martingale doubles your stake after every loss, so the first win recovers everything plus £1. Here's what 10,000 sessions of each approach produced:
| Result | Martingale | Flat £1 bets |
|---|---|---|
| Reached £50 profit | 75.84% | n/a |
| Lost the whole £200 | 24.08% | 0.0% |
| Finished ahead | 75.89% | 24.98% |
| Average result | -£10.24 | -£13.60 |
| Worst session | -£200 | -£96 |
| Average spins played | 102.7 | 500 |
The Martingale felt like a winner three times out of four. Then, roughly once in four sessions, a run of losses pushed the required stake beyond what was left, and the whole budget went. Averaged across every session, it still lost money, because the house edge applies to every pound staked, and the Martingale stakes a lot of pounds. Flat betting lost a little more per session on average, £13.60 against £10.24, but it played five times as many spins to do it, so each spin cost far less. It also never lost more than £96 in a session, while the Martingale's bad sessions took everything. The bankroll calculator lets you try your own stakes.
How often does a slot session finish in profit?
It depends heavily on volatility, even when the RTP is identical. All three model slots below pay back exactly 96%, and each ran 100,000 sessions of 200 spins at £1.
| Model slot | Spins that pay | Sessions ahead | Median result | Middle 80% | Best session |
|---|---|---|---|---|---|
| Low volatility | 48.6% | 38.76% | -£11 | -£54 to +£42 | +£180 |
| Medium volatility | 25.7% | 39.43% | -£25 | -£92 to +£98 | +£475 |
| High volatility | 18.2% | 21.41% | -£86 | -£126 to +£82 | +£2,191 |
The average losses, £8.03, £7.94 and £6.75, sit close to the £8 that a 4% edge on £200 of spins predicts. The high-volatility figure moves around more because a handful of huge wins pull its average. But the typical experience was very different. On the high-volatility slot, the median session lost £86 of the £200 staked, and only about one session in five finished ahead, while a lucky few landed wins of more than £2,000. That's why the same RTP can feel generous on one game and brutal on another. The slot volatility calculator shows the same effect with your own numbers.
What does this mean for how you play?
Three things stand out. First, patterns in past results mean nothing, whether it's a hot number, a colour streak or a bonus round that feels overdue. Second, betting systems reshape your results without improving them, and the reshaping usually means rare but heavy losses. Third, volatility decides what a session feels like far more than RTP does, so match the game to your budget, not the other way round.
None of this needs AI to work out. It's arithmetic, repeated enough times to see the pattern clearly. That's also why every "AI predictor" for these games fails: there's nothing in the results for it to find.
Can you use or share this data?
Yes. The data and code are free to reuse under a Creative Commons Attribution licence (CC BY 4.0). Just credit ai-casino.online and link to this page. Journalists, students and forum posters are welcome to use the charts and figures.
- Roulette number counts (CSV, 37 rows)
- Martingale and flat betting sessions (CSV, 20,000 rows)
- Slot session results by model (CSV)
- Full summary (JSON) and the Python code
A note on limits: these are simulations of fair games with fixed rules, not recordings from real casinos. Real slots have their own pay tables and volatility, and real players change their stakes, which the flat and Martingale sessions don't.
What should you do next?
Put the findings to work with your own numbers.
FAQ
What else do people ask about these simulations?
Does roulette even out over time?+
Closely, but never exactly. Over one million simulated European roulette spins, every number landed within about 1.5% of its expected count of roughly 27,027, which is what a fair wheel produces.
Is red more likely after several blacks or reds?+
No. In one million simulated spins, red came up 48.62% of the time straight after five reds in a row, the same as its normal rate of 48.65%. Each spin is independent.
Does the Martingale system work in roulette?+
It wins small amounts most of the time and loses large amounts occasionally. In 10,000 simulated sessions with a £200 budget, 76% reached £50 profit, 24% lost the full £200, and the average session lost £10.24.
How often do slot sessions end in profit?+
In our simulation of 200-spin sessions on 96% RTP model slots, 39% finished ahead on low volatility, 39% on medium and 21% on high.
Can I use this data?+
Yes. The data, charts and code are free to reuse under CC BY 4.0 with credit and a link to ai-casino.online.
Keep exploring
More casino AI tools & calculators
Tools that use this research
Other games and tools