Slot RTP and House Edge: The Arithmetic, Not the Numbers

Wilhelmina Coote / Clause-by-clause editor

Our library holds a hundred operator cards with thirteen fields each, and not one of those fields is a return percentage. There is no slot RTP figure anywhere in the data, on any card, for any game. That is worth saying at the top rather than burying, because most pages on this subject print figures they cannot source. RTP is published per game and our cards are keyed to operator domains, so the two never meet. What we can do instead is set out the arithmetic properly, mark every illustrative number as a placeholder, and then show which fields our cards do hold that actually bear on a session. The arithmetic is simple. The sourcing is the hard part, and it is the part nearly everyone skips.

Where a slot RTP figure is not: our hundred cards

The field list on a card runs as follows: an identifier, a display name, a domain, a founding year, a licence object, a coin list, a verification object, a limits object, a restricted-country list, the address that list was read from, a tag list, a note on whether the pages were read live or from an archived copy, and the date they were read. Thirty-three cards add a studio list. Three add a note on location masking. One adds a payout object.

No return percentage, no volatility band, no game count, no release date. The thirty-three studio lists hold 548 distinct name strings between them and not one of those strings carries a figure. So there is no slot RTP column to publish, and inventing one would be the same failure as inventing a verification figure or a payout time. The thirteen fields we do carry were chosen because each one can be traced back to a page and a date.

That is a deliberate limit rather than an oversight. An RTP is set by the studio that built the game, is sometimes configurable by the operator who licences it, and is published per game rather than per operator. Our cards are keyed to operator domains, so even a complete set of per-game figures would not sit anywhere on them. The register lookups behind our licence lines are indexed by domain too, which is the same reason.

What follows is therefore arithmetic and structure, not measurement. Every number used as an example below is labelled as a placeholder at the point it appears.

The arithmetic: an edge is what is left over

Write R for a published RTP percentage. The house edge is whatever R does not account for, so the edge as a percentage is 100 minus R. That is the whole relationship, and it is an identity rather than an estimate.

• If R is 96, the edge is 4. Both figures are arithmetic placeholders, not readings from any game.
• If R is 94, the edge is 6. Same caveat.
• If R is 99, the edge is 1. Same caveat.

The identity holds whatever the source of R, which is the only pleasant property in this whole subject: you do not need to trust anybody to perform the subtraction, only to trust the figure you subtract from a hundred. Two consequences follow that are worth more than any specific figure. First, a small change in RTP is a large change in the edge: moving R from 96 to 94 is a two-point move in the return and a fifty per cent increase in the edge, because 6 is half again as much as 4. A table that ranks games by return figure alone makes that look like a minor difference, and in edge terms it is not.

Second, an edge is a long-run proportion of turnover and not a prediction about any session. It describes what the arithmetic tends towards across a very large number of outcomes, and the number of outcomes required for the tendency to show is far larger than a single sitting. It says nothing about what a hundred spins will do, and any page that converts an edge directly into an expected evening is misusing it.

So the honest form of a slot RTP claim has three parts: the figure, the game it belongs to, and where the figure was published. Drop any one of the three and the number becomes decorative. We hold none of the three, which is why this page carries arithmetic and no readings.

The edge multiplies turnover, not your deposit

This is where most explanations go wrong, and it is pure algebra, so it can be set out exactly.

Let s be the stake per spin, n the number of spins, and e the edge as a decimal. Total turnover is s times n. The long-run expectation of what the house retains is s times n times e. Notice what is not in that expression: the amount deposited. Turnover is the quantity being multiplied, and turnover grows every time a returned stake is staked again. A balance of a hundred can generate a turnover of several thousand without a second deposit, which is why deposit size is the wrong number to reason from.

A worked example with placeholder numbers throughout, none of them read from a game: at s equal to 1, n equal to 500 and e equal to 0.04, the product is 20. If the same balance is recycled so that turnover reaches 5,000 rather than 500, the same edge produces 200. The edge did not change. The turnover did. Both 0.04 and the spin counts are placeholders chosen to make the arithmetic legible, not figures read from any game.

The practical reading is that turnover, not deposit size, is the quantity a house edge applies to, and turnover is the one variable a player controls directly. That is an arithmetic statement about the formula rather than a suggestion about how much anyone should stake, which is not something a document can tell you and not something we would guess at.

Where the operator documents do publish numbers you can divide, they are caps and deadlines rather than returns, and those are set out in the divisors the documents do publish.

Why the studio name is the nearest thing we hold

If a return percentage cannot be read off our cards, the studio list is the closest stand-in, and it is a weak one. It tells you which companies wrote the games in a lobby, which is at least the level at which an RTP is set.

Counting across the 33 lists, Endorphina and Pragmatic Play each appear on 26, Hacksaw Gaming on 25, PG Soft on 24, then BGaming, GameArt and NetEnt on 23 apiece, Evolution on 22, and AvatarUX and Betsoft on 21. At the other end, 264 of the 548 strings appear on one list only.

Two warnings attach to those counts. They are string counts, so one company can appear several times under different spellings: Betsoft appears as Betsoft on 21 lists and Betsoft Gaming on 3, Hacksaw as Hacksaw Gaming on 25 and Hacksaw on 7, Nolimit City in three spellings across 14, 6 and 3 lists. And a name on a list says nothing about how many titles sit behind it, which may be one or four hundred.

So the studio column narrows the question from the operator to the supplier and then stops. It cannot be turned into a slot RTP figure by any route, and we do not score it at all: it has a nominal weight of one point out of ten and is switched off as unscored on every card, because there is no register of studios to check a name against. The full count is in which studios appear on the most cards.

A return figure says nothing about how a session moves

Two games can share an RTP and behave entirely differently, because the figure is an average and averages discard shape. One game may return in many small amounts, another in rare large ones, and the same R covers both.

That property has its own name and its own page, and it is genuinely independent of the return figure: see why two games with one return figure feel different. The point for arithmetic is that an average and a distribution answer different questions, and a single percentage only answers the first.

Our cards hold neither. No field records a volatility band and no field records an RTP, so on this site both properties are described structurally rather than quantified. That is a limit of our data, stated plainly, not a claim that the properties do not matter.

A slot RTP ranking is therefore the clearest example of something we will not publish, whereas a ranking of operators on what they do publish about the account is fair game. The distinction runs through the whole site: game-level facts we cannot verify stay in prose, and account-level facts with a clause number behind them go into the score.

What the cards do carry that bears on a session

Nineteen numbers across a hundred cards, and all of them concern the account rather than the reels.

Fourteen cards state a withdrawal cap: Bitcasino.io 1,000,000 USDT per week in clause 6.10, CoinCasino 500,000 EUR per month in clause 9.6, Wild.io 100,000 USD per week in clause 9.6, Oshi Casino 4,000 EUR per day in clause 12, Instant Casino 4,000 EUR per day in clause 8.1, Slotum 15,000 EUR or USD per week in clause 13, CryptoLeo 50,000 USDT per month in clause 7.9, Mirax 10,000 EUR per week in clause 8, PlayAmo 7,500 EUR per week in clause 12, BetPlays 5,000 EUR per week in clause 6.11, Bitdreams 20,000 EUR per month in Section 11, Wild Fortune 15,000 EUR per month in clause 11.8, Rocketpot 5,000 USD per month in clause 11.5, and Bets.io 10,000 USD per transaction in clause 10.

Five cards state a verification figure: 2,500 EUR at Bitcasino.io, Sportsbet.io and Livecasino.io, each in a clause 6.6; 2,500 USD at Rocketpot in clause 11.4; 2,000 USDT at Empire.io in clause 5.4. Two cards state a deadline in days, at fourteen and thirty. One card, Vave, states a three-day maximum for processing in clause 8.1 and instalments over up to thirty days above 50,000 USDT in clause 8.8, read on 25 August 2026.

Every one of those is checkable by clause number against a page read on a stated date, ninety-nine of the hundred cards on 26 August 2026 and one on 2 September 2026. A slot RTP figure on a comparison table almost never carries that apparatus, which is the whole reason we prefer the nineteen numbers we can source to the hundreds we cannot.

Three questions to ask of any slot RTP figure

Since we cannot supply the figures, the next most useful thing is a test for the ones other pages supply. Three questions, and a figure that fails any of them is not worth carrying into a comparison.

Which game, exactly? An RTP belongs to a single title, not to an operator and not to a studio. A page that prints one percentage beside a casino name has attached a game-level number to an account-level object, and there is no arithmetic that makes that valid. The nearest honest version is a studio list, and we hold 33 of those covering 548 name strings.

Published where, and read when? Our own cards answer that question for every field they carry: ninety-nine were read on 26 August 2026 and one on 2 September 2026, thirty-eight from live pages, forty-one from an archived copy and twenty-one not at all. A slot RTP figure with no address and no date behind it cannot be rechecked by anybody, which means it also cannot be corrected.

Which configuration? The same title can ship in more than one setting, so a single figure may be one of several. That is not a claim about any particular game, it is a reason a bare percentage is an incomplete statement even when it is accurate.

Compare that with how a licence line behaves. OGL/2024/328/0599 can be taken to a register and returns a holder, a grant date and a status; ALSI-202509013-FI1 returns a holder company and an issue date. Sixty-eight of our hundred cards carry such a number, thirty-seven from the Curacao Gaming Authority and twenty-five from the Anjouan Gaming Board. Thirty-two carry none. An RTP has no equivalent register anywhere, and that asymmetry is the real subject of this page.

49,500 searches, and the figure nobody can source

The demand here is large and the sourcing is thin, which is an unhealthy combination. Keyword Planner, United Kingdom, 26-27 September 2026 gives 49,500 monthly searches for online slots and 27,100 for slot games UK, with online slots UK at 22,200.

Against that, an RTP is published per game, can differ between configurations of the same title, and is not something an operator card can carry. So the volume of the question and the availability of a sourced answer are badly mismatched, and the gap gets filled with figures whose origin nobody states.

Our position is the dull one. We print the arithmetic, which is an identity and cannot be wrong: the edge is 100 minus R, and it multiplies turnover rather than deposits. We print the placeholders as placeholders. We print the nineteen account-level numbers with their clause references. And we leave the slot RTP cell empty, because an empty cell is a fact about our reading and a fabricated figure is not a fact at all.

If you came here for a ranked list of games, we do not have one and will not assemble one from figures we cannot check. If you came to compare operators on what they actually publish, the cards we could read are ranked among the crypto casinos in the UK on our front page, the field-by-field survey is at which fields an operator fills in, and the game names behind the search volume are collected in the slot names readers search for.