Chips & Truths No spin. Just the math.

The Mathematics of Deterministic Randomness

A long-form analysis of PRNGs, virtual reels, RTP structure, near-misses, and the deterministic architecture behind modern slot algorithms.

By H. Omer Aktas

The Mathematics of Deterministic Randomness: PRNGs, Virtual Reels, and the Architecture of Modern Slot Algorithms

I. Abstract

Modern slot machines are often described as “random,” but the word can obscure more than it explains. Electronic Gaming Machines are not little mechanical drums waiting for luck to settle on a symbol. They are regulated mathematical systems. A random selection process chooses an outcome from an approved distribution; a paytable assigns value to that outcome; and the screen converts the result into reels, sound, motion, anticipation, and loss-or-win feedback.

This paper examines the mathematical architecture behind that process — certified random number generation, virtual reel mapping, and expected-value accounting through the PAR sheet. The decisive historical shift was the move from physical reels, where probability was constrained by the number of stops on a reel, to virtual reels, where the displayed symbols can be separated from the true probability distribution. Telnaes’s 1984 patent sits at the center of this transition: it describes a machine in which physical reel positions are displayed as the result of an electronically generated random number, rather than serving as the complete probability structure themselves (Telnaes, 1984).

In a standard independent-outcome EGM, the machine is not “due,” not compensating, not remembering, and not moving toward a win. Each play is drawn from the approved mathematical model. The short run may feel patterned because human perception is highly sensitive to streaks, near-misses, and apparent rhythm. The underlying mathematics is indifferent to all of it. What appears to be mechanical suspense is the display of a result already selected by software.

II. From Mechanical Reels to Mathematical Reels

The old mechanical slot machine had a fixed constraint: its mathematics was tied to its body. If a reel had 22 stops, and each stop was equally likely to land on the payline, the probability of any one stop appearing was:

$$P(s)=\frac{1}{22}$$

For a three-reel machine with one jackpot symbol on each reel, the top-award probability was:

$$P(\text{jackpot})=\frac{1}{22}\times\frac{1}{22}\times\frac{1}{22} =\frac{1}{10,648}$$

In a purely mechanical game, visible reel geometry and mathematical probability were closely connected. A player couldn’t necessarily calculate the game, but the reel itself contained the mathematical structure. If the manufacturer wanted a rarer jackpot, the machine needed more stops, more reels, fewer jackpot symbols, or some other physical compromise.

This created a design ceiling. Very rare prizes are commercially attractive because they support large advertised jackpots. But a mechanical machine cannot easily offer million-dollar-style odds while still looking like a compact, familiar three-reel slot. Making the jackpot rare enough would require an impractical number of stopping positions.

The microprocessor solved that problem by changing what the reel was. The reel became a display surface, not the source of probability. The machine could still look like a traditional slot, but the true probability model moved into software. Telnaes’s patent described precisely this break: a random number is chosen from a set larger than the number of physical reel stops, with one physical position represented by one or several virtual positions (Telnaes, 1984). The visible reel remained, but its mathematical authority was reduced.

The machine does not spin and then discover the outcome. The outcome is selected first, and the spin displays it.

III. Randomness in a Deterministic Machine

Casino machines are not “random” in the folk sense of being mysterious, moody, or uncontrolled. They use certified random selection processes that must satisfy technical standards. In many EGMs, the random source is a software pseudo-random number generator. Some standards also permit hardware random number generators, provided they meet monitoring and statistical requirements (Gaming Laboratories International, 2016; Nevada Gaming Control Board, 2016).

A software PRNG is deterministic at the machine level. If the algorithm, internal state, and seed were exactly reproduced, the same sequence would follow. That does not make it predictable to the player. A properly designed and certified gaming RNG is seeded, advanced, protected, and tested so that its outputs are statistically suitable for game play and practically inaccessible to outside prediction.

A simplified transition model is:

$$X_{t+1}=F(X_t)$$

Here, $X_t$ is the internal state at time $t$, and $F$ is the transition function that produces the next state. The player does not control $F$, cannot observe $X_t$, and cannot know which future state would correspond to a favorable result.

Pressing the spin button samples an outcome at a moment in time. It does not create a favorable condition. If the RNG is advancing rapidly, even a microscopic timing difference produces a different sampled value, but the player has no information that would allow that timing to be exploited. “Timing the button” is not a strategy.

Nevada Technical Standard 1 gives a useful picture of how regulators think about this. It requires, among other things, that a software RNG not use a static seed, that it cycle at a minimum average rate of 100 Hz, and that it not draw RNG values for future play (Nevada Gaming Control Board, 2016). The regulatory intention is straightforward: the outcome must be available, random within the approved design, and protected against prediction or outside influence.

A machine that is deterministic internally can still be random for gambling purposes. Most practical randomness in computing works this way. The relevant question is whether the game outcome is unpredictable, statistically valid, independent where required, and faithful to the approved game mathematics.

IV. Virtual Reels and Symbol Weighting

The virtual reel is the key to modern slot mathematics. It allows the game designer to separate what the player sees from the true probability of what the player can receive.

Suppose a reel displays a small set of visible symbols:

$$S={s_1,s_2,\ldots,s_m}$$

Now suppose the software reel contains $N$ virtual stops. Each virtual stop maps to one displayed symbol:

$$\phi:{1,2,\ldots,N}\rightarrow S$$

If $w(s_i)$ is the number of virtual stops assigned to symbol $s_i$, then the probability of that symbol is:

$$P(s_i)=\frac{w(s_i)}{N}$$

with:

$$\sum_{i=1}^{m}w(s_i)=N$$

A symbol can appear important, central, or visually frequent while still carrying a much lower probability on the payline than the player assumes. The probability is not governed by the symbol’s visual weight. It is governed by its virtual-stop count.

A simplified virtual reel might look like this:

SymbolVirtual StopsProbability
Blank3232/64 = 50.00%
Cherry1818/64 = 28.13%
Bar1010/64 = 15.63%
Seven33/64 = 4.69%
Jackpot11/64 = 1.56%

When the RNG selection maps to virtual stop 42, the machine checks the mapping table and displays the symbol assigned to that stop. The player sees a reel stop on a symbol. The software has selected an entry in a table.

The reel animation is not a transparent probability diagram. It is a representation of a weighted mathematical system. Telnaes’s invention made it possible to keep the familiar appearance of a reel machine while allowing odds no longer limited by the number of physical stops (Telnaes, 1984).

V. Outcome Space, Paytables, and Expected Value

For a game with $k$ reels, each reel may have its own virtual reel size $N_j$ and its own symbol weighting. The outcome space is:

$$\Omega={1,\ldots,N_1}\times{1,\ldots,N_2}\times\cdots\times{1,\ldots,N_k}$$

If the reel selections are independent, the probability of a particular virtual-stop combination $\omega=(r_1,r_2,\ldots,r_k)$ is:

$$P(\omega)=\prod_{j=1}^{k}\frac{1}{N_j}$$

For displayed symbols, the probability becomes:

$$P(s_1,s_2,\ldots,s_k)=\prod_{j=1}^{k}\frac{w_j(s_j)}{N_j}$$

The paytable then assigns awards to outcomes. Let $A:\Omega\rightarrow \mathbb{R}_{\geq 0}$, where $A(\omega)$ is the award for outcome $\omega$. If the wager is $b$, theoretical return to player is:

$$RTP=\frac{\sum_{\omega\in\Omega}P(\omega)A(\omega)}{b}$$

The house advantage is:

$$HA=1-RTP$$

So if a machine carries a theoretical RTP of 92%, the house advantage is:

$$HA=1-0.92=0.08$$

and the player’s expected result per unit wagered is:

$$E[\text{player profit}]=RTP-1=-0.08$$

That number is often misread. It does not mean the player loses exactly 8% on a session, and it does not mean the machine must return money after a losing run. It means that over a sufficiently large number of plays, the weighted average of all possible awards is 92% of the amount wagered.

The PAR sheet documents this design. PAR sheets contain the probability and accounting structure of the game: reel strips or virtual reels, symbol weights, award frequencies, hit frequency, theoretical hold, and related mathematical data. Harrigan and Dixon’s study of Ontario slot-machine PAR sheets found that games looking similar to players can have meaningfully different payback percentages and structural characteristics (Harrigan & Dixon, 2009). The visible cabinet is not the mathematics.

VI. Why a Machine Is Not “Due”

The belief that a machine is “due” is one of the most durable errors in gambling. It feels reasonable: if a game is programmed to return a certain percentage and it has not paid for a long time, the mind wants to complete the pattern. It imagines a hidden balancing mechanism somewhere inside the cabinet.

That is not how standard independent-outcome slots work.

Let $J_t$ be the event that a jackpot occurs on spin $t$. Under independence:

$$P(J_{t+1}\mid J_t^c,J_{t-1}^c,\ldots,J_{t-n}^c)=P(J_{t+1})$$

The probability of a jackpot on the next spin is unchanged by the fact that previous spins lost.

The mistake comes from confusing long-run expectation with short-term correction. The RTP is real, but it is not a promise to any individual player or session. It is the mean of a distribution. The law of large numbers says observed averages tend toward expected averages as the number of trials becomes large. It does not say that a particular player, machine, or session will be compensated for a cold streak.

Nevada Regulation 14 addresses this directly. Gaming devices must theoretically pay out a mathematically demonstrable percentage, all possible game outcomes must be available at the initiation of play, and devices must not alter any function based on actual hold percentage (Nevada Gaming Commission & Nevada Gaming Control Board, Regulation 14). A machine that tightens, loosens, remembers, or self-corrects after a streak would be in violation of that standard. Cold and hot are player labels, not machine states.

VII. Near-Misses: The Most Persuasive Losing Result

The near-miss deserves separate treatment because it sits at the collision point between mathematics and psychology. When a jackpot symbol appears just above or below the payline, the player feels close. Physiologically, the body reads it as almost-success. Mathematically, the outcome is a loss, no different from any other losing combination.

In a virtual reel system, visual adjacency does not imply mathematical proximity. A symbol appearing just above the line is there because the display sequence shows that arrangement — not because the RNG came within some numerical distance of a jackpot. The RNG selected a value mapping to a losing result, and the animation displayed it in a way the human eye reads as close.

Clark, Lawrence, Astley-Jones, and Gray found that near-miss outcomes in a simulated gambling task increased motivation to continue gambling and recruited win-related brain circuitry, even though the near-miss paid nothing (Clark et al., 2009). Near-misses can therefore carry motivational weight that the mathematics does not support.

The behavioral evidence is not uniform, however. Pisklak, Yong, and Spetch reviewed the near-miss literature and found that evidence for its effect on continued gambling behavior is mixed under tightly controlled conditions (Pisklak et al., 2020). The stronger claim — that every near-miss automatically drives continued play — is not well supported. What holds up better is that near-misses can increase perceived closeness, emotional arousal, and motivation in certain contexts, with behavioral effects depending on design and measurement method.

For anyone at the machine, the practical point is straightforward: a near-miss does not change the next spin. It is a losing outcome with persuasive presentation.

VIII. Stop Buttons, Timing, and the Illusion of Control

Many players believe that stopping the reels manually changes the outcome. In most modern games, the outcome is determined when the play begins. The stop button shortens the animation. It does not alter the result.

The confusion is understandable because the machine responds: a button is pressed and the reels stop. The action looks causal. But a display responding to input does not mean the probability was changed by it.

Even if the button press could identify the exact instant of RNG sampling, the player has no access to the RNG state, no way to predict the next value, and no method for selecting a favorable position. There is no skill to apply.

Harrigan and Dixon noted that stop buttons and similar features contribute to the illusion of control in slot-machine play (Harrigan & Dixon, 2009). The player is given interaction, but not influence. Approved skill-based or hybrid games, where player decisions can genuinely affect outcomes, must be disclosed as such under relevant regulatory standards. In a standard independent-outcome slot, button timing, hand position, stopping style, and personal ritual carry no mathematical weight.

IX. RTP, Volatility, and the Shape of Experience

RTP describes the long-run expected return. It says nothing about how the game feels day to day.

Two games can both carry a 92% theoretical return while behaving very differently. One may pay frequent small awards; the other may concentrate its return in rare bonus rounds or top prizes. The expected value is the same. The variance is not.

$$Var(X)=E[X^2]-(E[X])^2$$

A high-volatility machine produces long dry periods interrupted by large hits. A low-volatility machine delivers more frequent small awards but fewer dramatic events. Neither pattern confirms whether a game is fair. They are different shapes of the same expected-value structure.

Personal session experience is therefore a poor diagnostic. A player sitting at a high-volatility machine for an hour may feel nothing is happening. Someone who arrives later may hit a bonus quickly. Neither outcome proves timing, destiny, or machine memory. Short-run samples from a high-variance distribution are inherently unstable — that instability is part of the design.

The casino’s long-run advantage does not require every session to look smooth. It requires only that the approved distribution, repeated over many wagers, contains a negative expectation for the player.

X. Operational and Regulatory Significance

For the casino operator, the EGM’s mathematics lives in the approved configuration. Theoretical hold comes from the paytable, virtual reel mapping, and outcome distribution. No manager decides which player should win, and no machine releases money on schedule. The edge is built into the design before the machine reaches the floor.

For the regulator and testing laboratory, the question is whether the device implements the approved design faithfully. That covers the RNG, seeding, scaling, game logic, paytable, program authentication, meters, logs, and resistance to outside interference. GLI-11 provides a broad technical framework for gaming-device evaluation, including software and hardware RNGs, seeding, and monitoring (Gaming Laboratories International, 2016). Nevada Technical Standard 1 gives jurisdiction-specific requirements for random selection processes and RNG integrity (Nevada Gaming Control Board, 2016).

Anecdotal suspicion is genuinely difficult to evaluate, and that difficulty is partly structural. On a large floor, rare events are statistically normal. Long losing streaks, back-to-back bonuses, sudden jackpots, and clusters of near-misses will all appear somewhere because the number of plays is enormous. The individual witness sees one dramatic local event. The approved model accounts for the full distribution.

A properly regulated slot machine does not need to manipulate individual spins. The house edge is already there.

XI. Conclusion

The modern slot machine is a probability engine in the costume of a mechanical reel game. Its screen preserves the old language of spinning reels, stopping symbols, and near-wins. Its mathematics is something else: random number generation, virtual reel mapping, paytable-weighted outcomes, expected value, variance.

The central change is the separation of appearance from probability. On a mechanical reel, visible stop positions largely defined the odds. On a modern EGM, the reel is only a display layer over a deeper mathematical table. The player sees motion; the machine has already selected a result.

That architecture explains why the common myths fail. A machine that has not paid is not building pressure for a future payout. A jackpot symbol just above the line is not mathematical progress toward anything. A stop button gives the player interaction, not control. A high RTP does not guarantee a good session; a losing streak does not prove manipulation.

The game may feel mechanical, emotional, and personal. Its mathematical foundation is none of those things. It is a static approved distribution, executed independently, one wager at a time.

References

Clark, L., Lawrence, A. J., Astley-Jones, F., & Gray, N. (2009). Gambling near-misses enhance motivation to gamble and recruit win-related brain circuitry. Neuron, 61(3), 481–490. https://doi.org/10.1016/j.neuron.2008.12.031

Gaming Laboratories International. (2016). GLI Standard #11: Standards for Gaming Devices, Version 3.0. Gaming Laboratories International.

Harrigan, K. A., & Dixon, M. (2009). PAR sheets, probabilities, and slot machine play: Implications for problem and non-problem gambling. Journal of Gambling Issues, 23, 81–110. https://doi.org/10.4309/jgi.2009.23.5

Nevada Gaming Commission & Nevada Gaming Control Board. Regulation 14: Manufacturers, distributors, operators of inter-casino linked systems, gaming devices, new games, inter-casino linked systems, on-line slot metering systems, cashless wagering systems, mobile gaming systems, interactive gaming systems and associated equipment.

Nevada Gaming Control Board. (2016). Technical Standards for Gaming Devices and Associated Equipment: Technical Standard 1, Integrity of Gaming Devices.

Pisklak, J. M., Yong, J. J. H., & Spetch, M. L. (2020). The near-miss effect in slot machines: A review and experimental analysis over half a century later. Journal of Gambling Studies, 36, 611–632. https://doi.org/10.1007/s10899-019-09891-8

Telnaes, I. S. (1984). Electronic gaming device utilizing a random number generator for selecting the reel stop positions. U.S. Patent No. 4,448,419.

Play smart. Gambling involves real financial risk. If the game stops being entertainment, it's time to stop playing.