RTP Pillar·Chapter 7

RTP vs the returns you actually see

A 96% RTP is a long-run average, not a promise for your afternoon. Any finite session is a single draw from a wide distribution around that mean — here is why the return you actually see rarely matches the published figure, and what your own results can and cannot tell you.

By Vadym Zelinskyiverified 2026-09-18Current · 100%

Read a slot’s information screen and it gives you one number: a return-to-player of, say, 96%. It is easy to hear that as a rate you should roughly see — put in a hundred, expect ninety-six back. That is not what the figure means, and the gap between the published RTP and the return a real session produces is not a defect. It is the entire subject of this chapter. The published number is a long-run expectation; what you experience over an afternoon is a single sample drawn from a wide distribution around it.

An expectation, not a promise for your session

RTP is an average over an enormous number of rounds. Formally it is the expected value of the return per unit staked, computed from the game’s model — see how RTP is calculated. An expected value describes where the average of many independent rounds settles as the count grows without bound. It says nothing about any particular round, and very little about any particular session. A finite stretch of play has its own distribution of outcomes; 96% is only the centre of that distribution, not a floor, a ceiling, or a prediction for the next hour.

Why sessions wander: variance and the law of large numbers

The reason a session strays from 96% is variance, and the reason it strays less as you play more is the law of large numbers. The realized average return converges to the expected value only as the sample grows, and the rate of that convergence is precise: the standard error of a sample mean is the per-round standard deviation divided by the square root of the number of rounds — the s/√N standard-error term that sits inside the standard confidence interval for a mean. The spread of your observed return shrinks in proportion to 1/√n, no faster. Small n means a wide spread, so your realized return can land far from 96% in either direction; only a very large n squeezes it close.

“Very large” here is larger than intuition allows. Because the spread falls as 1/√n, cutting your uncertainty in half requires four times the data. Our companion chapter turns this exact arithmetic into a calculator: pinning a medium-volatility slot’s return to within ±0.1% at 95% confidence takes on the order of 96,040,000 spins — roughly 111 days of continuous play at about ten spins every second, without pause. That figure is the flip side of this chapter: the same 1/√n law that makes long-run RTP verifiable at all is what makes a single session almost uninformative about it. Turn the crank yourself in how RTP is verified.

Same RTP, different ride: volatility

Two games can share a 96% RTP and feel nothing alike, because RTP fixes the average but leaves the shape of the distribution open. That shape is volatility: how often a game pays anything at all (hit frequency) and how the win sizes are spread when it does. A low-volatility game pays small amounts often, so sessions stay near the average; a high-volatility game pays rarely but can pay large, so most sessions run below the average while a few rare ones sit far above it. Standard deviation is the usual measure of this spread — Wizard of Odds describes it plainly as “a measure of how volatile your bankroll will be”, with roughly 68% of outcomes falling within one standard deviation of expectation and about 95% within two. The published RTP alone tells you where the distribution is centred and nothing about how wide it is — which is why two games at an identical return can offer completely different session experiences.

The house edge is the certainty; your session is the variance

State it plainly: over enough play, the mathematics favours the house. A 96% RTP is a 4% house edge, and that edge is the deterministic part — the direction the long-run average is guaranteed to move. The variance is the random part, and in the short run it dominates the edge completely, which is exactly why a winning session is entirely possible. But possible is not the same as favourable. The distribution a session is drawn from is centred below break-even, so variance is not an opportunity hiding inside the edge; it is noise around a negative mean. It cuts both ways — some sessions finish up, more finish down — and averaging over all of them lands on the house’s side by construction. A good night does not bend the expected value; it is just a draw from the favourable tail of a distribution whose centre never moved.

Reading your own results honestly

So what can a player read from their own numbers? Almost nothing about the game itself. A session return is a single, high-variance sample: it estimates the game’s RTP with an uncertainty so wide that it is compatible with almost any plausible RTP. You cannot estimate a game’s return from a session, and you certainly cannot detect a one-percentage-point difference between two games from anything short of millions of rounds — the 1/√n arithmetic forbids it. What your results do honestly tell you is what happened to your bankroll: how much you staked, how much is left, and how the session felt. Those are facts about your afternoon, not measurements of the machine.

A run far below 96% is not evidence a game is “broken,” and a run far above it is not evidence a game is “loose” — both are ordinary points in a wide distribution. Nothing in a past session shifts the next spin: each round is drawn afresh from the same model, so no result is ever “due.” The only reliable statement your numbers support is a description of the session that already happened.

Read this way, the published RTP and your realized return stop contradicting each other. One is the centre of a distribution measured over tens of millions of rounds; the other is a single draw from that distribution over a few hundred or a few thousand. They are not supposed to match, and the mathematics that separates them is the same mathematics that lets the number mean anything at all.

Common questions

If a game’s RTP is 96%, why did I lose far more than 4% of my stake?

Because 96% is the long-run mean, not what any single session returns. A finite run of spins is one draw from a wide distribution centred on that mean, and over a short session the spread of that distribution is far larger than the 4% edge itself. Losing much more — or much less — than 4% across one session is ordinary variance, not a sign the number is wrong.

Can I work out a game’s real RTP from my own results?

No. The precision of an estimate of the mean return improves only in proportion to 1/√n, so halving your uncertainty means quadrupling the number of rounds. Pinning a return to within a tenth of a percent takes on the order of tens of millions of spins, as the spins-to-verify chapter shows. A session — even a long one — is nowhere near enough data to estimate a game’s RTP, let alone detect a fraction-of-a-percent difference between two games.

I had a winning session — doesn’t that prove the game pays above its RTP?

No. A winning session is exactly what variance produces some of the time; it is expected, and it proves nothing about the underlying return. Over enough play the mathematics favours the house, so the expected value of continued play stays negative regardless of how any one session ended. A good result is a draw from the same distribution as a bad one.

Does a lower-volatility game give me a better chance of ending ahead?

Volatility changes the shape and spread of session outcomes, not their average. At the same RTP, a lower-volatility game clusters results more tightly around the (negative) expected value and a higher-volatility game scatters them more widely; neither makes the expected return positive. Lower volatility narrows the ride in both directions — it does not tilt the edge in the player’s favour.

Sources (3)

Education, not advice. This chapter explains how return-to-player is defined, computed and checked so you can read the number honestly. It is not a system, and nothing here treats gambling as a way to make money — over enough play the mathematics favours the house. 18+.

End of the current pathBack to the RTP pillar