Cluster pays slots: the grid arithmetic behind them

Take away paylines, take away the left-to-right rule, and what remains is a grid where a win is simply a group of matching symbols touching each other. Cluster pays slots are that idea, and the whole behaviour of the design follows from two countable things: how many positions the grid has, and how many of them touch each other. A seven-by-seven grid has 49 positions and 84 touching pairs. Both numbers are arithmetic you can verify on squared paper in a minute, and this page works through what they imply. As always, our files hold no game titles, no grid sizes and no return figures, so every grid below is a substitution and is labelled as one.

The counting rule, stated exactly

In a line design, the game declares a set of position sequences and checks symbols along each. In a ways design, the game counts every selection of one position per reel, left to right. In a cluster design it does neither. It looks for groups of identical symbols that are adjacent — usually meaning orthogonally adjacent, sharing an edge rather than a corner — and pays a group once it reaches a stated minimum size.

Three properties fall out of that immediately. Direction is irrelevant: a cluster in the bottom-right corner pays exactly as one starting at the left edge. Reels are irrelevant as an organising idea; the grid is a field of positions, not a row of strips. And group size is continuous in a way a line win is not — a cluster of eight is a different payout from a cluster of seven, so the pay table is indexed by size rather than by symbol count along a line.

That third property is the real difference. In cluster pays slots the pay table has to state a value for every group size from the minimum up to the whole grid, which means the designer is setting a curve rather than a handful of line values. How steeply that curve rises between, say, size six and size twelve is the single most consequential decision in the design, and it is never published in marketing copy. The product rule this design discards is worked out in the product rule that cluster designs discard.

Positions and neighbours: arithmetic you can check

Substitute a square grid of side n. The number of positions is n². The number of orthogonally touching pairs is 2n(n − 1), because there are n rows each containing n − 1 horizontal pairs, and n columns each containing n − 1 vertical pairs.

Now break the 7 × 7 down by how connected each position is. Four corners have two neighbours each. Twenty edge positions have three each. Twenty-five interior positions have four each. Check the arithmetic: 4 × 2 + 20 × 3 + 25 × 4 = 8 + 60 + 100 = 168, and 168 ÷ 2 = 84 pairs, because every pair was counted from both ends. The figures agree, which is the point of showing them: every number quoted about the geometry of cluster pays slots can be reconstructed from the grid alone, and a figure that cannot be reconstructed did not come from the grid.

The consequence is geometric and worth holding onto. Just over half the positions on a 7 × 7 grid — 25 of 49 — are fully connected on four sides, and those are where clusters most easily grow. A matching symbol landing in a corner has two chances to extend; one landing in the middle has four. This is why cluster designs feel concentrated toward the centre of the grid, and it is not a design flourish, it is 4 versus 2.

The minimum cluster size is the main dial

Every cluster design states a minimum: below it, a group pays nothing. Move that minimum by one and the whole behaviour of the game moves, far more than changing the grid size would.

Substitute a 7 × 7 grid with s equally likely symbol types and independent positions. The expected number of positions holding any one particular symbol is 49 ÷ s. With s = 7 that is 7 positions of each type per spin; with s = 10 it is 4.9; with s = 12 it is about 4.1.

Set the minimum at five. With s = 7, the average spin puts seven of a given symbol on the grid, and a qualifying cluster needs five of those seven to be mutually touching — plausible but far from automatic, since seven positions scattered across 49 cells more often form two or three small islands than one connected block. With s = 12, the average spin puts only four of a given symbol on the grid, so a five-cluster requires an above-average count before connectivity is even a question.

So the symbol count and the minimum size interact multiplicatively, and the grid size is the least important of the three. That ordering — minimum size, then symbol count, then grid — is the useful mental model for cluster pays slots, and it comes entirely out of the arithmetic above. None of the numbers is a reading: our files record no grid, no symbol count and no minimum for any game, from any studio, anywhere.

Why cluster pays slots cannot work without cascading

Nearly every cluster design removes the winning group and drops new symbols into the vacated positions, then checks again. That is not decoration. Without it, a cluster win would occupy positions that can no longer contribute, and the grid would simply sit there.

The arithmetic of the chain is where the large outcomes come from. Suppose, as a substitution, a cluster of six clears on a 7 × 7 grid. Six new symbols arrive into positions that are now surrounded by whatever survived, and those six arrive adjacent to existing symbols by construction — a cleared region is, by definition, a hole with a rim. The probability of a second win on the refill is therefore not the same as the probability of a win on a fresh grid; it is higher, because the refill lands into an already-sorted neighbourhood.

That effect is bounded and it is why chains end. Each clear removes one symbol type from a region, so the refill is less likely to match the rim it landed against as the chain continues. The result is a distribution of chain lengths with a long thin tail: mostly one or two steps, rarely many.

A designer sets the average return against the whole chain, not against the first check. This is the reason a chained design can show a run of five consecutive wins that still totals less than the stake, and it is the honest explanation for why cluster pays slots feel generous while returning whatever the pay table says they return. Our files hold no chain data for any game, so the mechanism above is described structurally and the substituted cluster of six is an illustration.

What our files hold about the studios, and what we refuse to add

Thirty-three of our hundred operator records list the studios named on their own lobby. Counted after merging spelling variants, out of 33:

Here is the refusal, and it is deliberate. No field in our library connects a studio to a mechanic. We are not going to tell you which of the names above build cluster grids, because we would be filling that in from recollection and presenting it beside figures that are readings. The table says which suppliers are switched on across these lobbies and nothing else.

Two counting notes that change the numbers if ignored. Peter & Sons appears as both Peter & Sons et Peter and Sons; ELK as both ELK et ELK Studios; AvatarUX as AvatarUX et Avatarux; Print Studios as Print Studios et Printstudios. Across the whole library there are 548 distinct provider strings, and a substantial share of them are second spellings of a studio already counted. A raw string tally of cluster pays slots suppliers would report several studios that do not exist and undercount every one that does. The full census is in the full supplier census behind this table.

The pay table sets the return; the grid sets the feeling

It is worth stating the general result once, because it applies to every mechanic and is repeatedly ignored. A game’s return is the sum over outcomes of probability times payout. Any change to the counting rule changes which outcomes exist and how likely each is; it does not change what the designer chooses to pay for them. Grid size, minimum cluster, symbol count and cascade behaviour are all inputs to the probability side. The pay table is the other side, and it is set last.

So two designs with identical grids and identical minimums can return quite different fractions of stake, and two designs with wildly different grids can return the same. Nothing about the geometry constrains the return, which is why no arithmetic on this page produces a percentage. Our files contain no return-to-player figure for any game from any studio, and a percentage borrowed from elsewhere cannot be traced, so none appears here — which is why geometry constrains variance and not return.

What the geometry does constrain is variance, and the constraint is real. A larger grid with a low minimum produces frequent small clusters. A smaller grid with a high minimum produces long droughts and occasional grid-filling outcomes. Anyone comparing cluster pays slots is comparing those shapes, and the shapes are visible from the stated minimum and grid alone — which is more than can be said for the return. Pricing a feature under the same blind spot is covered in pricing a feature when the return is unpublished.

Where a cluster win stops being a screen event

A grid-filling outcome is the selling point of these designs, so the clauses that govern large balances are the relevant half of the comparison. Fourteen records in our library publish a payout cap, read from the operator’s own terms with the clause number recorded.

Read a substituted win of 60,000 units against the caps we did read. Oshi Casino caps payouts at €4,000 a day in clause 12: fifteen payout days. Instant Casino, €4,000 a day in clause 8.1: fifteen. BetPlays, €5,000 a week in clause 6.11: twelve weeks. PlayAmo, €7,500 a week in clause 12: eight weeks. Slotum, €15,000 a week in clause 13: four weeks. Bitdreams, €20,000 a month in Section 11: three months. CryptoLeo, 50,000 USDT a month in clause 7.9: two months. Bets.io caps a single transaction at $10,000 in clause 10: six transactions. Wild.io, $100,000 a week in clause 9.6: one.

Vave’s section 8 adds the other pattern: a maximum processing period of three days in clause 8.1, and balances above 50,000 USDT paid in equal instalments over up to thirty days in clause 8.8, with a fourteen-day document period in clause 8.7. 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. The clauses are transcribed in the Vave section 8 clauses read line by line.

None of those figures has anything to do with grids. They are the arithmetic that decides what a cluster filling a 49-position board is worth in a bank account, they are written in numbered clauses, and they can be checked in the time it takes to read a marketing page about the mechanic. They are also what builds the top no KYC casinos whose terms these clause numbers come from.

Three sentences sum up everything this page asserts. The geometry of cluster pays slots is fully derivable — 49 positions, 84 touching pairs, 25 fully connected cells on a 7 × 7 board — and needs no source beyond arithmetic. The return is not derivable at all, and our files hold no figure for it. The caps and clause numbers above are readings, dated, and they are the part of the comparison that can actually be settled.