The mathematics inside a reel machine

The mechanical constraint and how it was removed

An early mechanical reel machine had genuinely countable odds. Each reel carried a fixed number of physical stops, every stop was equally likely, and the probability of any combination was the product of the probabilities on each reel. Three reels of twenty stops give eight thousand equally likely combinations, and a single specified combination occurs once in eight thousand spins.

That arrangement caps the top prize. If the largest payout is funded by a combination occurring once in eight thousand spins, it cannot be much larger than a few thousand times the stake without pushing the total return above one. Adding physical stops to allow rarer combinations makes the reels physically larger, and the mechanism sets a limit on how far that can go.

The virtual reel removed the constraint. The machine holds an internal list of positions that is much longer than the number of stops shown on the physical reel, and several entries in that list map to the same visible stop. Selection happens uniformly over the internal list, so the visible stops are no longer equally likely: a stop appearing many times in the list is common, and one appearing once is rare.

Weighting, and why the strip cannot be counted

The consequence is that counting the symbols on a reel strip tells an observer nothing about the odds. Suppose a reel shows twenty stops and the internal list holds two hundred and fifty-six entries. If the top symbol is assigned a single entry, its probability on that reel is one in two hundred and fifty-six rather than one in twenty. Across three such reels the top combination occurs once in 256 x 256 x 256, which is 16,777,216 spins.

A top prize funded at that frequency can be very large without endangering the return, which is the entire purpose of the arrangement. It also means the machine can display the top symbol on adjacent positions frequently while the winning line remains extraordinarily rare, since the weighting applies to which stop is selected, not to what is printed next to it.

Working out the return

The return calculation is the same one used everywhere else in this category. For each paying combination, multiply the probability of that combination by the amount it pays, and add the products. The result is the expected return per unit staked, and one minus that figure is the house edge.

Because the probabilities come from the weighted list rather than the visible strip, the calculation cannot be performed by anyone who only has the machine in front of them. The designer knows the list, so the designer knows the return exactly; the player sees the paytable, which gives the payouts but not the probabilities. The two halves of the calculation are distributed asymmetrically, and that asymmetry is the single most consequential design fact about the device.

It is also the reason that testing and disclosure regimes exist at all. Where a wheel or a pair of dice can be inspected by anyone who understands counting, a weighted machine can only be verified by examining its internal tables, which requires access that no player has. The entry on regulation follows what was built in response.

Random selection and the meaning of a cycle

The selection itself is made by a generator producing numbers with no usable pattern, mapped onto the internal list. Two properties matter: the numbers must be uniformly distributed across the list so the weighting means what it says, and successive selections must be independent so no sequence of past results narrows the next one.

Independence has a practical consequence that is frequently misstated. The published return is a limit reached over the machine's full theoretical cycle, which for the weighting above runs to many millions of spins. It is not a promise about any hour, day or machine, and a device that has paid nothing for a long stretch is not thereby closer to paying. The generator has no memory, exactly as the die in the first entry of this board has none.