What we read inside casino lobbies, market by market

RTP data collection in numbers: markets, lobbies, days, disagreements

Below is the size of the collection as of the last recount of the showcase file. Every figure is recomputed by script from the lobby snapshots themselves; not one is typed by hand. Next to each headline number stand the figures you cannot read it without: how many lobbies refused, how many went unread that day, how many calendar days the window spans and how many days were missed.

Collected on 36 days between 2026-08-12 and 2026-10-09: 59 calendar days, 23 of them missed. Counts recomputed on 2026-10-09.

The large number counts the lobbies we managed to read on the last day their market was collected. Lobbies that refused, and lobbies we could not reach that day, stand beside it as separate numbers: what was not measured is never written as zero and is never added to refusals. Measurement records whose snapshot did not save which lobby answered stand as their own numbers, split the same way — read, refused, not read: what is missing is the name, not the state. A lobby read on two markets is counted once.

When each market was last read

Cadence is not the same everywhere, and no market is read every day: some are read weekly, others less often, and a market is only read from a vantage point where its lobbies answer. The date below is the last day that market was actually read.

The «Shopfronts on this market» column shows what the sweep covered that day: how many shopfronts were read, how many refused, and how many we did not ask at all. A refusal and «not asked» are different things, and neither is recorded as zero. Measurement rows for which the snapshot did not record which shopfront answered are shown separately as «unidentified», and by state as well: they cannot be counted as read, because two such rows may be the same shopfront, but their state must not be lost either — what is missing is the name, not what happened. The shopfronts themselves are not named: which operators we poll per market is the recipe, not the result.

The «Read from» column is the exit label we INTENDED to read from. Next to it is the result of the check: geolocation services are asked where the request actually left from. «Unconfirmed» means the services disagreed — the data was collected, but we do not claim where from. «No check» means an older snapshot that did not carry this check at all.

What we measure, and why it is not available anywhere else

Nobody publishes which RTP build a casino installed

A studio certifies several builds of the same slot and publishes the list. That part is open. Which of those builds a given casino installed is published by nobody: not the studio, not the operator, not the regulator. It is visible only in the lobby itself, and only on the day you look.

One snapshot shows today; a run of snapshots shows the history

So we look repeatedly, market by market. One reading tells you what is installed right now. A run of readings tells you when it changed, and that is the part nobody can back-fill. Cadence differs by market and each market below carries its own last-read date; a day we did not collect is a gap we cannot go back and fill, so the gaps are listed rather than smoothed over. History before the day you started collecting does not exist anywhere to be bought.

Every number on this page is recomputed from the snapshots themselves. None of them are typed in by hand.

Terms this page uses

Short definitions, each linked to the full entry in our glossary.

RTP, return to player

The share of stakes a slot machine returns to players over a very long run. It is a property of a certified build, not of your session.

House edge

The mirror of return to player: whatever is not returned over the long run stays with the casino. We do not collect it separately; it follows from the published figure.

Volatility

How lumpy the payouts are. Two builds with the same return play very differently, and lobbies print this field far less often than they print the return.

Slot machine

A game for which the studio certifies one or several return builds. Slots make up the bulk of what we read on casino shelves.

Payline

The pattern a winning combination is counted along. It is not part of our shelf-fact count: lobbies publish it less often than the studio or the shelf name.

Jackpot

A shared prize pool the lobby announces next to a game. Whether a game takes part in a draw is one of the shelf facts we count.

Progressive jackpot

A jackpot that grows with stakes and resets after a win. For us it is the same shelf fact: the lobby either announces the game takes part, or it does not.

Variance

The spread of outcomes around the expectation. It has nothing to do with lobbies disagreeing: that is different builds, not randomness.

What the shelves publish besides return-to-player

Return-to-player is a narrow good. Only some suppliers encode a build version, only some operators print a number, and every figure has to be proven before it can be used.

The shelf itself is wider and harder to argue with. Operators publish which studio made each game, which shelf it sits on, whether it feeds a jackpot — and, on some lobbies, the date the game went live on that shelf. We checked that this date describes the shelf rather than the game: lobbies running on shared technology report it identically, while independent ones disagree on roughly every second title.

That matters because it is history we did not have to wait for. Our own observation runs to 36 days — and some lobbies publish the shelf-arrival dates themselves, from long before that window.

How fresh these figures are

Lobby-published history reaches back to 2017-02-07

The date is what the operator states, not what we measured. We have shown it varies between lobbies; we have not shown it is accurate. That check becomes possible only once our own record is long enough to compare against.

These counts are a union across shopfronts, each taken from the last day it was actually read: 2026-08-20 at the earliest, 2026-10-09 at the latest. A shopfront we could not reach on the most recent collection day is shown from its previous reading rather than counted as empty. Of these, 35 shopfronts have not been read for more than 10 days: their most recent reading still counts here, inside the date range just given, rather than being dropped to zero.

Where companies disagree

Taking each market at its most recent reading, 352 games carry different certified returns at different companies in the same market. Two lobbies inside one company do not count, that is a deployment detail. A disagreement is two different companies holding two different builds of the same game. The last-read date of every market is listed above.

Markets shown here are the ones where a disagreement was found on the last observed day. The full collection covers 15 markets.

Why we print both directions

Both directions are printed on purpose. A company running the higher build in one market runs the lower one in another, and the reverse holds too. This describes how the market is configured. It is not an accusation against anyone in the table.

Published work you can cite

• Slot RTP versions 2026

  The frozen dataset behind the studio side of this work: which games ship in more than one certified version. Carries a DOI and an open licence.

• Methodology

  Sources, snapshot chain, confidence level per field. Read this before citing anything we publish.

• Public API

  JSON endpoints over the published registry. No key required.

• RTP editions by country

  Where the same game ships as different certified editions, and how that looks market by market.

• Slot RTP registry

  The public registry of published returns per game, every row carrying its source URL.