Return to player is a model's average, not a session promise

Return to player is a property of the game, not of your session

Every real-money game is built on a defined set of outcomes with defined chances attached to them. From that set, the designer can work out what fraction of everything staked the game gives back on average if it is played a theoretically unlimited number of times. That fraction is what return to player means. It is derived from the rules of the game in the same way the chance of a particular dice roll is derived from the shape of the die, and it exists before anyone has played a single round.

What it is not is a forecast about you. An average over an immense number of rounds says nothing in particular about a few hundred, in the same way that a known long-run split between heads and tails says nothing about the next ten flips. The figure is a correct statement about a model and a useless statement about an evening, and almost every misunderstanding on this subject comes from reading the first as though it were the second. It is also entirely separate from the cashier: it governs what happens at the game, while whether money can leave is a question about the route out of the account.

What the figure is calculated over

The arithmetic is done over the whole outcome space, weighted by how likely each outcome is. For a wheel or a card game that space is small enough to write down. For a modern slot it is vast: every reel stop combination, every path through a feature, every retrigger, every way a feature can be entered, each with its own probability. Where the space is too large to enumerate directly, it is established by simulating rounds in volumes that run to the billions.

That scale is the point, and it is also the catch. A figure that only becomes visible across billions of rounds is not something a person encounters. On slots played for real money, a single player's lifetime is a rounding error against the sample the figure is drawn from. The number is honest and the number is irrelevant to your afternoon, and both of those statements are true at once.

Why you cannot measure it from inside the game

Suppose you decide to check the claim. You stake, you record, you divide what came back by what went in. The trouble is that your result will differ from the stated figure, often by a lot, and that difference tells you nothing, because the expected spread of a short sample around a long-run average is wider than the thing you are trying to detect. A session well above the figure and a session well below it are both completely ordinary.

Nor can you reason backwards from what you saw. A long run of nothing is not evidence that a game has been turned down, and a feature landing twice in quick succession is not evidence it has been turned up. Both are what randomness looks like close up. The practical consequence is that the stated figure has to be taken from the documentation rather than from experience, and that anyone claiming to have measured it over a few sessions has measured noise.

There is a second reason the test cannot work, and it is about what the figure is averaged over. A large part of many games' return is carried by outcomes that occur very rarely — a top feature, a maximum win, a long multiplier chain. A player who never meets one of those is sampling only the common part of the distribution, which sits below the overall average by design. So the typical session result is not merely noisy; it is systematically under the headline, and that is the model working as intended rather than evidence against it.

Volatility is what the average hides

Two games can share a stated return and feel nothing alike, because the average says what comes back and not how it comes back. Volatility is the shape of the distribution: whether the return arrives as frequent small amounts or as rare large ones. A low-volatility game pays often and modestly; a high-volatility one pays seldom and, when it does, substantially.

This is the figure that actually governs how a session feels, and it is why identical stated returns produce entirely different experiences. A grid game built around a feature loads much of its return into rare events, so long dry stretches are a designed property rather than a malfunction. A cluster-paying release distributes differently again. Many games now state a volatility rating alongside the return, and when the two are read together they tell you far more than either does alone.

Where a game states its own figure

Every game has an information panel, reached from a menu, an i button or a paytable link inside the game itself, and that panel is where the number lives. It is written by the studio that built the game, it applies to the exact build you are about to play, and it is the only source that is both specific and current. A figure quoted in an article, a list or a review is a copy of something, and copies go stale.

It is worth getting into the habit of opening that panel before staking, because the same panel carries everything else that matters: how the features trigger, what the maximum win is expressed as, how the paylines or clusters work, and the volatility rating. On a release where the feature is the whole game the panel is the only place the structure is set out properly. The same applies to a single-line release such as a classic-styled title, where the simplicity of the screen hides the maths behind it.

One game, several versions

Here is the fact that breaks most quoted figures. A studio can configure one title at more than one return and license those configurations to different operators, so the game with the name you recognise may exist in several builds with different stated figures. The artwork, the sounds, the features and the name are identical. The maths underneath is not.

Which build you are playing is decided by the operator's licensing arrangement, not by you, and the only way to know is to open the information panel in the client you are actually using. That is precisely why no honest guide can print a number for a named game: the number belongs to a build, and the guide does not know which build you have in front of you. It is also why two people can both be right while quoting different figures for the same title.

Nor is the build necessarily fixed for ever. A studio can revise a title, an operator can move between configurations when a licensing arrangement changes, and a game that opened in a demo client is not guaranteed to be the same configuration as the real-money one beside it. None of that is hidden — it is simply only knowable at the point of play, from the panel, in the client you are logged into. Treat the figure as something you check each time you open an unfamiliar title rather than something you learnt once.

House edge says the same thing from the other side

House edge is the return figure subtracted from the whole. If a model gives back a certain share of stakes over the long run, it retains the rest, and the retained share is the edge. The two are one statement expressed twice, so a game cannot have a generous return and a large edge at the same time.

Convention decides which you see. Table games are usually discussed in terms of edge, and machines in terms of return, which makes the two look like different concepts when they are arithmetic complements. On roulette played for real money the edge follows directly from the layout, and the number of pockets on the wheel against the payouts offered changes it for every bet on the table at once. That is a rare case where the mechanism is simple enough to inspect yourself.

A payout percentage in advertising is a different number

Marketing occasionally quotes a site-wide payout percentage, and it is not the same quantity as a game's return. A site-wide figure is a backwards-looking ratio: everything paid out across everything staked, aggregated over some period and some set of games. It describes what happened to a crowd. A game's return describes what a specific model does by design.

That makes the advertised version easy to move without lying. Change the period, change which games are included, change whether bonus play counts, and the ratio moves with it. It also says nothing about the game you intend to open, because the mix of games other people chose is doing most of the work. When the two are set side by side the difference becomes obvious, and it is the reason a site-wide figure is interesting at best and never actionable.

What testing houses actually check

Games sold into the licensed British market are tested before they go live, but it helps to be precise about what the testing is for. An independent laboratory examines the random number generator, confirms that the implemented game matches the mathematical model the studio submitted, and verifies that the stated return corresponds to that model. In other words, it checks that the description is true of the software.

It does not audit your account, follow your session, or certify that you personally will see anything in particular. Nor does testing make a game favourable; a tested game with a modest return is still a modest return, correctly described. The assurance on offer is accuracy of description and integrity of the draw, which is worth having and is not the same as a promise about results.

Return to player and bonus turnover

The two interact, and the interaction is usually understated. A turnover condition requires you to stake a multiple of something, and it specifies which games count towards that total and at what rate. Games with higher stated returns are frequently the ones discounted or excluded, precisely because they would clear a condition too cheaply.

So the return figure is not the only variable in play when a bonus is open. A game that contributes at a reduced rate makes the turnover longer in stake terms whatever its return, and a stake ceiling while the bonus is live caps how fast you can get through it at all. Reading those together is the only way to understand what a bonus is asking of you, and the turnover rules themselves are where the contribution table is set out.

Live games are described differently from slots

A dealt or spun game has no reel maths, so its return is a property of the bets available on it rather than of the game as a whole. One table can carry several bets with quite different arithmetic, and a rules panel will often state a figure per bet rather than one for the table. That distinction matters more than it sounds: choosing a different bet on the same table changes the maths entirely.

Game shows complicate it further, because side bets, multipliers and bonus rounds each have their own structure. On a wheel-based game show the published figures are usually given per bet type, which is the honest way to present it. The same is true across live dealer tables generally: look for the rules or help panel on the specific table, not for one number covering the category.

A winning session proves nothing about the figure

It is tempting to treat a good run as confirmation that a game is paying well, and a bad one as evidence it has been tightened. Neither inference works, because both are drawn from samples far too small to carry the conclusion. The stated figure is a limit that the long run approaches, and nothing about a short run is required to resemble it.

This cuts both ways, which is the part worth holding on to. A good session does not mean the game owes you nothing further, and a bad one does not mean it owes you anything. There is no ledger between you and the game that evens up later, and no state the game carries from one round to the next unless its own rules say so explicitly. In Britain a player's winnings are not taxed, so the tax position on a win at least does not complicate the arithmetic further.

What to read before you stake

Four things, in the game's own panel rather than anywhere else: the stated return for the build in front of you, the volatility rating beside it, the maximum win however it is expressed, and how the main feature is triggered. Those four together describe the game far better than any single number, and all four take about a minute to find.

Then hold the figure in proportion. It is a correct description of a model, not a rate of pay, and no arrangement of it turns gambling into a source of income — a game designed to return less than it takes cannot do otherwise on average. Decide what you are prepared to stake before you open anything, treat it as spent, and keep that decision separate from whatever the panel says. If the figures are becoming a reason to continue rather than a description of a product, the pages on stepping back set out the tools, and free confidential help is at begambleaware.org or on 0808 8020 133. Whether the games you are playing are tested to the British standard at all depends on the operator holding a licence, which is public record at gamblingcommission.gov.uk.