Slot RTP and volatility: the arithmetic, not the number

There is no return-to-player figure anywhere in our fact files. Not one, for any game, from any studio, on any of the hundred operators we hold records for. So this page cannot tell you the slot RTP of anything, and it is not going to borrow a percentage from a comparison table and present it as a reading. What it can do is the arithmetic: how a stated return becomes an expected loss, why turnover rather than stake is the multiplier, what volatility adds as a second and almost never published number, and how many spins it would take to detect a house edge by playing. Every worked figure below is a substitution, and each is labelled as one.

What a stated slot RTP is, and what our files hold instead

A stated return is a long-run average: of every unit staked across an enormous number of trials, the fraction returned to players. It is a property of a specific game in a specific configuration, and the same title can ship in more than one configuration, which is why a percentage without a game name and a configuration describes nothing in particular.

Our library holds none of them. What it holds, for 100 operator records checked on 26 August 2026, is: 33 lists of the studios named on a site’s own lobby, 49 summaries of verification wording, 27 of those with a clause number, 5 with a stated monetary figure, 38 country lists, 14 published payout caps, 30 coin lists, and licence entries checked against two public registers keyed by domain. Those are the numbers on this site. A return percentage is not among them, so none appears.

This is a deliberate refusal rather than an oversight. A slot RTP figure that has passed through three comparison sites has no chain behind it: nobody in the chain read it from a primary document, the configuration was dropped somewhere along the way, and the decimal place that survived is false precision. Printing it would make every other figure on the page less trustworthy by association, because a reader cannot tell from the page which numbers were read and which were inherited.

What follows is therefore arithmetic. Arithmetic needs no source: you can check every line of it on paper, and the substituted inputs are flagged so nothing here can be mistaken for a measurement.

From a stated return to an expected loss

If a game returns a fraction R of everything staked, the house edge is 1 − R. Substituting R = 0.96 gives an edge of 0.04, or four units lost per hundred staked. That is the entire conversion, and it is the first place most writing goes wrong, because four per cent sounds like a small number and is not being applied to what readers assume.

It is applied to turnover, not to the money brought to the table. Substitute a stake of 1 unit and 600 spins in an hour — a rate chosen to make the arithmetic visible, not observed anywhere. Turnover is 600 units. Expected loss is 600 × 0.04 = 24 units an hour.

Two readings of that table matter. The difference between a 0.96 and a 0.94 design is not two per cent of anything a player notices; it is a 50% increase in expected loss per unit staked, 36 against 24. And the whole column scales with turnover, so doubling the stake or the spin rate doubles it. A slot RTP difference that looks trivial in the headline is multiplied by every unit that passes through the game. A headline that looks like value and is not is taken apart in what a headline combination count does not say about value.

Volatility is the second number, and it is the one that is missing

A stated slot RTP tells you the average. It says nothing about the spread, and the spread is what a session actually consists of. The quantity that describes it is the standard deviation of the result per unit staked — call it s — and it is essentially never published.

Substitute two designs with identical returns of 0.96 and per-spin standard deviations of 3 and 30. Over 100 unit spins, both have an expected loss of 4 units. The first has a total standard deviation of 3 × √100 = 30 units; the second, 30 × √100 = 300 units. The expected outcomes are the same and the range of plausible outcomes differs by a factor of ten.

That factor is the whole of what “high volatility” and “low volatility” mean, and it is why the descriptive labels on game pages are so unsatisfying: they compress a number that spans an order of magnitude into three adjectives. Our files hold no volatility rating for any game either, so we cannot rank anything on it, and the labels on lobby pages are not readings we have taken.

The interaction with return is the useful part. Two games with the same slot RTP can be completely different products, and two games with different returns can feel identical if their spreads differ enough to swamp the gap. Over a hundred spins, a 2-point difference in edge moves the expected result by 2 units while a spread of 300 units decides what actually happens. Over a hundred thousand spins, the reverse is true. Which regime a player is in depends entirely on the number of trials, which brings us to the next section. The counting rules that set that spread in the first place are surveyed in the counting rules that set a game’s spread.

Why a personal test cannot detect a house edge

This is the arithmetic that should end the genre of “I played it and the return was terrible”, and it can be done in four lines.

Substitute an edge of 0.04 and a per-spin standard deviation of 10 units on unit stakes. After n spins, the expected loss is 0.04n and the standard deviation of the total is 10√n. For the expected loss to stand clear of the noise at, say, twice the standard deviation, you need 0.04n ≥ 2 × 10 × √n. Divide both sides by √n: 0.04√n ≥ 20, so √n ≥ 500, so n ≥ 250,000 spins.

A quarter of a million spins. At the substituted 600 spins an hour, that is more than four hundred hours at one table, on one configuration of one game, with the stake held constant throughout. Reduce the edge being tested to 0.02 and the requirement quadruples to a million spins, because n scales with the inverse square of the edge.

Three conclusions follow, and all three are arithmetic rather than opinion. Nobody has personally verified a slot RTP figure by playing; the sample required is out of reach of a session, a week, or a reviewer’s afternoon. A run of results in either direction over a few hundred spins carries no information about the design. And any page reporting a tested return without stating a sample size in the hundreds of thousands has reported noise. We have run no such test, have no outcome ledger, and print no tested figure. The clearest case of a margin that cancels out of every choice a player can make is the round where the margin cancels out of every target.

Bankroll arithmetic: the number that is actually under a player’s control

Since the edge is fixed by the design and the spread is unpublished, the only quantity a player sets is how much money passes through the game. That is worth working through because it is the one calculation where the inputs are known.

Substitute a balance of 500 units and a unit stake. At 600 spins an hour and an edge of 0.04, expected loss is 24 an hour, so the expected time to exhaust the balance on the average path is about 21 hours. Raise the stake to 5 units and turnover rises fivefold: expected loss 120 an hour, expected duration about four hours. Nothing about the game changed; the stake did.

Now add the spread. With a per-spin standard deviation of 10 on unit stakes, the total standard deviation after 600 spins is 10 × √600 ≈ 245 units against an expected loss of 24. So in a single substituted hour the noise is ten times the average effect, which is exactly why short sessions feel unrelated to any stated return, and exactly why they are.

The practical arithmetic, then, is that turnover determines the expected cost and the spread determines whether any individual session resembles it. Both are arithmetic. Neither requires a percentage from anybody, which is fortunate, because the percentage is the one input nobody outside the studio can verify. The one number a lobby does publish is examined in the one published price in a lobby, and its limits.

Where the arithmetic meets the terms page

Suppose the spread does its work and produces a large result. What that result is worth is decided by documents we did read, and this is where our files carry actual figures rather than substitutions.

Fourteen records publish a payout cap. Substituting a 30,000-unit result: Oshi Casino caps payouts at €4,000 a day in clause 12, so eight payout days. Instant Casino, €4,000 a day in clause 8.1, eight. BetPlays, €5,000 a week in clause 6.11, six weeks. PlayAmo, €7,500 a week in clause 12, four weeks. Mirax, €10,000 a week in clause 8, three weeks. Slotum, €15,000 a week in clause 13, two weeks. Bitdreams, €20,000 a month in Section 11, two months. Rocketpot, $5,000 a month in clause 11.5, six months. Wild.io, $100,000 a week in clause 9.6, one payout. CoinCasino, €500,000 a month in clause 9.6, one.

Five records state a figure at which additional identity checks apply: Bitcasino.io €2,500 in clause 6.6, Sportsbet.io €2,500 in clause 6.6, Livecasino.io €2,500 in clause 6.6, Rocketpot $2,500 in clause 11.4, Empire.io 2,000 USDT in clause 5.4. And Vave’s terms set a maximum processing period of three days in clause 8.1, a fourteen-day document period in clause 8.7, and instalments over up to thirty days for balances above 50,000 USDT in clause 8.8 — terms read 25 August 2026, licence OGL/2024/1676/0905 held by Latcas B.V., status Active, granted 19 May 2025, read 2 September 2026.

Compare the two halves of this page. The slot RTP half is arithmetic on substituted inputs because the real inputs are unpublished. This half is readings, with clause numbers and dates, and it can be checked by anyone who opens the same terms page. That asymmetry is the argument for spending more attention on the second half than on the first, and the second half is collected in the payout figures we did read, clause by clause.

Why our rating formula has no return component

Our scoring file weights seven things: licence 2.5, verification clarity 2.0, publishing a country list 1.5, coin coverage 1.5, reach 1.5, no verification demanded at signup 1.0, and provider breadth 1.0 — with provider breadth marked unscored, so it contributes nothing. A record needs at least three scorable components before it receives any number, which is why 47 of 100 records carry a rating and 53 carry null rather than a low score.

Game return is not on that list, and the reason is the one this page has been making. Every component above reads a field that came from a primary document: a licence number that can be looked up in a register, a clause number that can be found in terms, a coin list that can be counted, a country list with the address it was copied from. A return figure has no such field, for any operator, so a component scoring it would have to be filled from somewhere unverifiable — and one unverifiable component contaminates every rating that includes it, because the arithmetic would no longer trace all the way down.

The same logic keeps volatility off the list. There is no volatility field, no rating of it, and no way to derive one from what we hold. Omitting both is what lets us print the arithmetic behind every score: each component carries its own note, so a figure on any page can be walked back to the field it came from.

Six checks, and what this page asserts

First, look for a game name and a configuration beside any percentage. Without both, the figure describes no object.

Second, look for the source. A primary document means the studio’s own paytable or a certification report; a comparison table citing another comparison table is not a source.

Third, convert to an edge before reacting. 0.96 and 0.94 differ by two points in the headline and by 50% in expected loss per unit staked, 36 against 24 on the substitution above.

Fourth, ask about turnover. The edge is applied to everything staked, not to the money brought along, so spin rate and stake size move the cost proportionally.

Fifth, distrust every tested figure without a sample size. Detecting a 0.04 edge against a spread of 10 needs about 250,000 spins on the arithmetic above; halving the edge quadruples that.

Sixth, read the cap and the clause. A 30,000-unit result is one payout at one operator and eight daily ones at another, and that is written down with a clause number.

What this page asserts, in full: our files contain no return-to-player figure and no volatility rating for any game; every return, spread, spin rate and sample-size figure above is a labelled substitution; and the caps, thresholds, periods and licence details are readings from operator documents, dated. Everything on the slot RTP side of the line is arithmetic anyone can redo; everything on the document side is a reading anyone can repeat. Both kinds sit side by side in the top no KYC casinos and the fields that actually carry numbers.