Skip to content
Insights

Slot Simulations: What They Can and Cannot Prove

By Smiling Ghost · 4 October 2026 · Slot Mathematics

Every slot studio runs simulations, and every maths report quotes them. “Simulated over a billion rounds” sounds like proof. Sometimes it is, sometimes it is not. This article shows what a simulation actually tells you, where it runs out, and how a modern outcome-table approach changes the picture.

About the numbers. The examples use “Game A”, the illustrative model from Slot RTP Explained: RTP exactly 95.00%, with a 1,000x prize that lands once in 3,636 spins. It is not one of our games. Simulations were run in Python/NumPy with a fixed random seed.

1. A simulation estimates; it does not measure

A plain Monte Carlo simulation plays the game many times with random numbers and averages the result. Because the result is random, a second run gives a different answer. We ran Game A 50 times at each size:

Rounds per run Lowest estimate Highest estimate Typical spread
10,000 66.20% 128.38% ±15.07 points
100,000 80.11% 108.09% ±5.18 points
1,000,000 91.23% 98.16% ±1.70 points

The true value is 95.00% in every row. Even a million rounds can miss it by several points for a game like this, because so much of its return sits in one rare prize. A simulation result is only meaningful together with its sample size and an idea of the game’s volatility.

2. Rare events may not show up at all

The rarer an outcome, the more rounds you need to see it even once:

Outcome Rounds simulated Chance it appears at least once
1,000x (1 in 3,636) 1,000 24.0%
1,000x (1 in 3,636) 10,000 93.6%
Max win, 1 in 10 million 100,000 1.0%
Max win, 1 in 10 million 1,000,000 9.5%
Max win, 1 in 10 million 10,000,000 63.2%
Max win, 1 in 10 million 50,000,000 99.3%

That is why the maximum win and other rare outcomes are usually checked analytically or by a targeted search, not by waiting for random play to find them.

3. Outcome tables: simulation as a source, weights as the truth

Some platforms, Stake Engine among them, do not run the game logic live for every bet. The studio generates a large set of complete rounds in advance, gives each round a weight, and the server picks a round at random in proportion to its weight. With this approach:

  • The RTP is exact. It is the weighted average payout of the stored table, calculated, not estimated. Sampling noise from section 1 does not apply to the published figure.
  • Rare outcomes can be included on purpose. The open-source Stake Engine Math SDK lets the designer request a share of rounds of a given type, for example rounds that hit the maximum win, so they exist in the table. The weights then set how often each one is actually picked.
  • The table is the game. Whatever is not in the table cannot happen. This makes the maths easy to audit: anyone with the table can recompute every number.

Stake Engine’s math requirements ask for 100,000 to 1,000,000 simulated rounds per mode for slot games, so that results are varied enough. Our own games use hundreds of thousands of rounds per mode.

4. What simulations cannot prove

  • That the code is right. A simulation faithfully repeats whatever the code does, including its bugs. A wrong payline gives a precise, reproducible, wrong RTP. The fix is an independent check: recalculating rounds with separate code. We do this for every game before release.
  • That the random number generator is good. Simulations use the developer’s own random numbers. The RNG behind real bets belongs to the platform, and a simulation says nothing about it.
  • That the player sees the right thing. The animation must show exactly the result the server picked. That is a matter of front-end testing and replays, not maths.
  • How the game feels. Two games with the same RTP and volatility can be fun or tedious. Numbers do not decide that; playtesting does.

Takeaways

  • Ask how big a simulation was and how volatile the game is before trusting its RTP to a decimal place.
  • Do not expect random simulation to find very rare outcomes; look for how they were checked.
  • Outcome tables with weights make the RTP exact and auditable, but they still need independent checks of the code that produced them.

Related: Maximum Win, Feature Frequency and RTP Contribution · How we build and check our maths.

Want to work with us?

Distribution, media or a collaboration idea: tell us what you have in mind.