They Don't Need to Cheat

Stop arguing about rigged games

The gambling industry does not need to cheat you, and the twenty-year argument about whether it does has been the best protection it ever had.

You know the accusations. The random number generator is tampered with. Big accounts get throttled. The machine knows how much you staked. These claims are almost impossible to prove, and that is precisely what makes them so valuable to the people they are aimed at. An operator can ignore a hundred documented regulatory findings and spend its entire public response demolishing the one claim nobody can substantiate. It wins the argument, and the real record goes unread.

So this piece concedes the point entirely. Assume every game is scrupulously fair. Assume every wheel is true, every card shuffled honestly, every slot certified. Nothing below depends on cheating, and nothing below rests on an anonymous source or a leaked document. Every figure links to a regulator's decision, a government inquiry, a peer-reviewed study, or the industry's own marketing material.

On those terms, here is what the record shows.

• A revenue model in which a small minority of harmed players supplies a large share of the profit, so that reducing harm and reducing revenue are the same operation.
• Machines engineered against named psychological mechanisms, documented by the designers themselves, with industry terms for each technique.
• The same slot deployed at different payout rates depending on which operator serves it, disclosed only in a panel almost nobody opens.
• Regulators in the UK and Australia issuing record penalties and declaring major licensees unsuitable to hold a licence at all.

A rigged casino is a solvable problem. You change operator, and you are done. What the evidence actually describes is not solvable that way, because nothing is malfunctioning. An industry whose profitability depends structurally on people who have lost control is not suffering from a defect. It is working. The harm is not a cost of the business. It is the business.

That is the case below, and the industry's own numbers make it.

The customers it cannot afford to cure

In France, 40.2% of gambling revenue comes from problem gamblers. In Québec, 31.6%. In Germany, 32%. One research team, one methodology, three countries, published in 2019. The contrast is the other half of the finding: those gamblers are 4.8% of players in France, 2.7% in Québec and 4.6% in Germany.

Hold those numbers still for a moment, because the industry would much rather you argued about whether the roulette wheel is weighted.

These are not outliers. A 2019 evidence review assembled more than twenty separate estimates across seven countries and twenty-three years. Different researchers, different decades, different regulatory systems, different games. The answer keeps coming back the same: somewhere between a fifth and a half of all gambling revenue is supplied by the minority of players who have lost control.

The most recent market data points the same way. In Australia, where the measurement is most continuous, 12.7% of people who gamble generate 48.5% of revenue — Roy Morgan, roughly 16,000 interviews, twelve months to March 2024.

GREO evidence review (2019), 22 estimates across 7 jurisdictions, 1996–2019

Online makes the concentration sharper. Research on internet gambling found that between 38% and 67% of the highest-revenue players met criteria for gambling problems, against 24% to 35% of everyone else. The more a player is worth to the operator, the more likely that player is to be in trouble.

Why this is the whole argument

Treat those percentages as a business input rather than a social statistic, and the conclusion is unavoidable.

If every problem gambler on earth recovered tomorrow and began playing within their means, this industry would lose somewhere between a quarter and a half of its global revenue overnight. Not a segment. Not a slice of the downside. A third of the business, give or take, in every country where anyone has bothered to measure.

This means harm reduction and revenue protection are not competing priorities that a responsible operator balances. They are the same quantity with opposite signs. Every deposit limit that works, every self-exclusion that holds, every player who stops chasing losses, registers directly as lost earnings. An operator can fund awareness campaigns, print helpline numbers and sponsor research, and none of it costs anything as long as it does not change behaviour. The moment it does, it shows up in the accounts.

No conspiracy is required to produce this outcome, and no rigged game. It follows from the revenue distribution alone, which the industry's own market researchers publish.

Addiction is a product specification

The techniques below are not alleged. They are published, named, measured and in some cases patented. The industry designs against documented psychological mechanisms, and the academic literature describing how has been in print for over a decade.

Losses disguised as wins

On a modern multi-line slot, you can stake one unit, receive back less than one unit, and watch the machine celebrate. Lights, rising music, coins. You lost money. The machine told you that you won.

Dixon, Harrigan, Sandhu, Collins and Fugelsang measured this in Addiction in 2010. Analysing all 259,440,000 possible outcomes of one commercial machine (Lucky Larry’s Lobstermania), they found that losses disguised as wins do not exist at all when one line is played, and rise to 18.4% of spins at fifteen lines, exceeding the 14.2% rate of genuine wins. In live play sessions, these fake wins occurred on 17.1% of spins against 15.6% real wins.

The physiological result is the part that matters. Measuring skin conductance, the authors found that losses disguised as wins produced arousal responses statistically indistinguishable from real wins, and significantly greater than losses. The player's nervous system registers a win. The player's bank balance registers a loss.

Dixon et al., Addiction (2010) · live play sessions, 40 participants

That gap is not a side effect of multi-line design. It is what multi-line design produces, and it scales with the number of lines the machine encourages you to play.

The near miss

Two jackpot symbols land, the third stops just short. Reviewing the neuroscience in The Neuroscientist in 2016, Murch and Clark set out the evidence that near misses recruit the same reward circuitry as actual wins, and increase the desire to keep playing, despite being losses.

The frequency of near misses on a modern machine is a configurable parameter. It is not what random reel positions would produce.

Time on device

In Addiction by Design (Princeton University Press, 2012), MIT anthropologist Natasha Dow Schüll reports fifteen years of fieldwork among machine gamblers and the people who design and sell the machines. Her central finding concerns the industry's own objective function.

The target is not making you lose, and not making you win. It is time on device — an industry metric, in industry language. Everything else follows: the speed of resolution, the absence of clocks, the ergonomics, the reinforcement schedule. Players describe the resulting state as "the machine zone," a dissociative absorption in which winning has stopped being the point.

Variable-ratio reinforcement

The underlying schedule has been understood since Skinner's work in the 1950s. Rewards delivered after an unpredictable number of responses produce behaviour that is slower to establish and dramatically more resistant to extinction than rewards on a fixed schedule.

It is the most persistence-inducing reinforcement pattern in the behavioural literature, and it is the exact payout structure of every slot machine on every casino floor and in every app. This is not a coincidence that researchers noticed afterwards. It is the design.

The same game does not pay the same

The slot you played last night may exist in three or four versions. Same name, same artwork, same sounds, same animations. Different payout rate. Which version you were served was decided by the operator, not the game studio, and not disclosed anywhere you were likely to look.

The clearest confirmation comes from inside the industry. Game studio BGaming states it plainly in its own published explainer: "Providers often release several RTP versions for the same game," and "the casino chooses which one to run based on margin targets or commercial terms." Its worked example is a single title listed at 96% on one casino and running a 94% build on another.

That is not an accusation made by critics. It is a supplier describing its own product line.

The difference sounds small. Over a player's lifetime turnover it is not. A player who cycles 100,000 units through a 96% machine expects to lose 4,000. The same turnover through the 94% build of the same machine expects to lose 6,000. Half as much again, for an identical experience on screen.

A note on the numbers circulating online

Claims that a specific majority of casinos deliberately deploy the lowest available build are widespread, and they are usually published by affiliate sites that are themselves paid by operators, with no sample, period or method attached. They are not used here. The documented facts are narrower and sufficient: multiple builds exist, operators choose between them, and the choice is disclosed only in a panel most players never open.

The asymmetry that matters

This is where a real distinction has to be drawn, and where much of the popular criticism of online casinos goes wrong by aiming at the wrong target.

	Can the player verify an individual result?	What protects the player
Provably fair originals	Yes, by recomputation	Cryptographic commitment published before the bet
Third-party slots	No	Certification paid for by the industry, plus the licence
Skin-gambling sites, 2016 era	No	Nothing, as the record shows

What protects the player, by game type.

On a provably fair game, the operator publishes a hash of its server seed before you bet, and reveals the seed afterwards. You recompute the result yourself. Nobody has to be trusted, and no result can be altered after the fact without the recomputation failing. The design is sound, and attacking it is a waste of a critic's credibility.

On a third-party slot, you cannot recompute anything. There is no commitment, no revealed seed, no arithmetic you can run. Your only protection is a testing laboratory paid by the industry and a licence that, as the next section shows, is often issued by a jurisdiction with minimal capacity to enforce it.

The question for slots is therefore not whether they are rigged. It is a simpler and more uncomfortable one: what, concretely, would prevent it, and who would find out?

What the regulators already found

Nothing in this section is an allegation. Each item is a finding by a statutory regulator or a government inquiry, published with reasons.

Britain: the largest penalty in the regulator's history

In March 2023, the Gambling Commission imposed a record £19.2 million package on three William Hill Group companies: £12.5 million against the operator of williamhill.com, £3.7 million against Mr Green, and £3 million against the business running 1,344 betting shops.

The findings are specific, and they are worth reading slowly, because each line is one real customer.

• A new customer spent £23,000 in 20 minutes, with no checks.
• Another spent £18,000 in 24 hours after opening an account, with no checks.
• A third spent £32,500 over two days after opening an account, with no checks.
• One customer lost £14,902 in 70 minutes without being identified as at risk.
• Another lost £54,252 over four weeks with no evidence of income obtained.
• A customer was able to place a £100,000 bet against a £70,000 credit limit, because no delay applied to limit increases.
• A customer staked £42,253 across 130 bets in three days without being flagged or contacted.
• 331 customers who had self-excluded with Mr Green were allowed to gamble with WHG International — the same group, a different brand.

On anti-money laundering, the Commission found large deposits accepted with no source-of-funds evidence: one customer lost £70,134 in a month, another £38,000 in five weeks, another £36,000 in four days.

The penalty came one week after a £7.2 million penalty against Kindred Group. These are not isolated operators at the fringe of the market. They are among the largest licensed businesses in Britain.

Note what the self-exclusion finding means. 331 people had formally asked to be stopped, and the controls failed. Of all the numbers in this article, that one is the hardest to read as an accident, because those customers had already identified themselves as the group described in the revenue section.

Australia: licensees found unsuitable to hold a licence

Successive public inquiries examined the country's two largest casino operators and reached the same conclusion: neither was suitable to hold the licence it held. The Bergin Inquiry in New South Wales reported on Crown in 2021, the Finkelstein Royal Commission in Victoria followed the same year, and the Bell Inquiry reached the equivalent finding on Star.

Separately, the financial-crime regulator AUSTRAC secured a A$450 million penalty against Crown Resorts in 2023 over money-laundering breaches, and subsequently pursued Star.

The United States: when the result itself was arranged

The skin-gambling boom of 2016 produced the clearest documented case of a fixed outcome, and it is worth stating precisely because it is the exception that defines the rule.

CSGO Lotto was promoted by two influencers who streamed their own winnings to millions of viewers while concealing that they owned the site. Trevor Martin was its president; Thomas Cassell its vice-president. The Federal Trade Commission took action and settled in its first case against individual social-media influencers. The complaint is public.

The lesson is not that gambling sites rig outcomes. It is how this one was caught: not by a player checking the maths, because no player could. It took a federal investigation. That is what happens in an unverifiable system when something does go wrong.

Where the industry chooses to be licensed

The operators serving the largest online markets are frequently licensed somewhere else entirely, most often Curaçao, and increasingly Anjouan. This is a choice, and it is worth being precise about what the choice buys, because the lazy version of this criticism is now out of date.

Curaçao reformed. The Landsverordening op de kansspelen took effect in December 2024, replacing the old master-and-sub-licence arrangement with direct licensing by the Curaçao Gaming Authority. According to the jurisdiction's legal guidance, operators must now appoint a Responsible Gaming Officer and an independent Compliance Officer, run age verification, behaviour tracking, cooling-off periods of 24 hours to three months, self-exclusion from one year to permanent, and deposit limits. Credit-based wagering is banned. AML regulations followed in April 2025, with customer due diligence thresholds, suspicious transaction reporting and annual independent audits.

On paper, that is a real regime. Anyone claiming Curaçao has no player-protection rules is describing the situation before 2024.

So the honest criticism is narrower, and harder to dismiss.

First, the economics are explicit. Corporate tax falls to between 0% and 2% where substance requirements are met, and there is no gaming tax at all. Annual licence fees total roughly EUR 47,000. Compare that with the gaming duties levied in the large consumer markets these operators actually serve. The licence is a tax position before it is anything else.

Second, enforcement capacity is the open question. The rules above are new, the authority is still issuing sector policies, and nobody yet knows what sustained supervision of hundreds of operators serving millions of players on other continents looks like in practice. Compare the UK Gambling Commission, which produced the £19.2 million package described above, with its investigators, its enforcement history and its published decisions.

Third, and most concretely for players: recourse. If a licensed British operator withholds your balance, there is an escalation path with teeth in your own jurisdiction. If an operator licensed 7,000 kilometres away closes your account during a withdrawal, your practical remedy is a complaint form in a jurisdiction with no connection to you, in a legal system you cannot access, over a sum that will never justify the cost of pursuing it.

That asymmetry is the product. Not the absence of rules, but the distance between the rules and anyone able to enforce them on your behalf.

The arithmetic the player never sees

Players track their balance. The house tracks turnover. That difference explains almost everything about why the experience of gambling and its outcome diverge so completely.

Expected loss has one formula

Your expected loss is not the money you deposited. It is every unit you ever staked, multiplied by the house edge, counted again each time the same money goes round.

This is why a player can deposit a modest sum, feel they are roughly breaking even for hours, and still be handing over a large amount. Recycled winnings are staked again, and the edge applies on every pass.

Most players have never seen their own turnover figure, although most platforms display it. It is the only number that predicts what the game will cost them.

No betting system can change this

This is the one part of the subject that is settled mathematics rather than contested evidence, and it closes off an entire genre of advice.

Doob's optional stopping theorem gives the result. If each individual bet has negative expected value, then every sequence of such bets also has negative expected value, regardless of stake sizes, ordering, or any rule for when to stop. Martingale, Fibonacci, d'Alembert, Labouchère, Paroli: all are covered by the proof. They do not need to be tested one by one.

What a progression system changes is not the total. It is the shape of the loss, and that is precisely why it feels effective.

Take the most common system: double up after a loss, targeting a 1.30x return at roughly 76% success. Each completed cycle nets a small profit. To recover, the stake must grow by a factor of about 4.3 at every step.

Consecutive losses	Required stake	Cumulative exposure
1	1.00	1.00
2	4.30	5.30
3	18.53	23.84
4	79.77	103.60
5	343.37	446.98
6	1,478.12	1,925.10
7	6,362.87	8,287.97

Computed from the progression above · 1.3026× target, 76% win rate.

Three consecutive losses occur roughly once in 72 cycles. Five, once in 1,256. Neither is rare: at a few hundred rounds an hour, a player meets the first within the hour.

Now the decisive comparison. Each winning cycle returns 0.30 units. Recovering the 446.98 lost at the fifth step therefore requires about 1,490 successful cycles. The failure arrives, on average, after 1,256. The collapse comes before the winnings that would have paid for it, and raising the bankroll only moves the wall back one step, at 4.3 times the cost.

So a progression system does not reduce losses. It converts frequent small losses into frequent small wins punctuated by rare total ones. The player remembers the wins, which are numerous, and codes the collapse as bad luck, which is singular. The house edge is identical throughout.

This is the most important thing in the article for an individual reader, because it is the mechanism by which an intelligent, numerate person concludes from direct personal experience that they have found a system that works.

What should be demanded instead

Every item below follows from the evidence above, and none of it requires proving that anyone cheated.

Verifiability as the default, not a feature. Cryptographic commitment works. An operator publishes a hash before the bet and reveals the seed after, and the player recomputes the result. This already exists on crypto-casino originals, and it was adopted because the skin-gambling scandals destroyed public trust. There is no technical reason it cannot extend to slots. The only obstacle is that nobody is requiring it.

Published payout rates, per operator, per game. Not the provider's headline figure. The build actually deployed, displayed where the player chooses the game rather than inside a panel. An operator that selects the 94% version of a title advertised at 96% should have to say so on the tile.

Turnover shown next to the balance. Every platform computes it. Almost none displays it prominently. A player who could see cumulative turnover alongside their balance would understand their position in seconds, and the formula in the previous section would do the rest.

Affordability checks that trigger before the loss, not after the investigation. The William Hill findings describe customers losing £23,000 in twenty minutes and £54,252 over four weeks with nothing interrupting them. The controls were absent at the moment they mattered and produced a fine years later.

Self-exclusion that actually excludes. 331 people asked to be stopped and were not. This is the simplest test of whether an operator's responsible-gambling programme is real, and it is auditable.

Regulation in the market where the player lives. A licence issued where the operator pays no gaming tax and the player has no standing is a tax arrangement, not consumer protection.

The conclusion

The debate about whether online casinos cheat has run for twenty years, consumed most of the public energy available for this subject, and protected the industry throughout. It is unresolvable by design, and while it continues, the documented material goes unread.

That material shows an industry that takes roughly half its revenue from the eighth of its customers who are already being harmed; that builds machines against named psychological mechanisms, measured and published by researchers, with physiological evidence that players' bodies register losses as wins; that selects which payout version of a game to serve you and discloses it almost nowhere; and whose largest licensed operators have been fined records sums and declared unsuitable to hold their licences by the regulators of two wealthy democracies.

None of this is hidden. All of it is in regulator decisions, peer-reviewed journals, royal commissions and the suppliers' own marketing pages.

They do not need to cheat. The published rules are already enough.

That is the finding, and it is considerably harder to answer than an accusation of rigging, because there is nothing in it to deny.

Method and sources

Standard applied. Every factual claim links to a regulator decision, a government inquiry, a peer-reviewed study, an evidence review, or a supplier's own published material. No claim rests on an anonymous source, a leaked document, an unverifiable allegation, or a secondary site repeating a figure without method.

What was deliberately excluded. Claims that random number generators are tampered with. Claims that stake size influences outcomes on provably fair games. Specific percentages of operators choosing the lowest payout build, which circulate widely but appear only in affiliate-funded content with no sample, period or methodology disclosed. These were checked and left out. Including them would have given the industry the one paragraph it could attack.

Limitations, stated openly. The revenue-concentration figures come from survey-based classification, and estimates across studies vary widely by product and country, from roughly 15% to 50%. The 48.5% headline figure covers problem and moderate-risk gamblers combined, not problem gamblers alone. The slot-design research on losses disguised as wins was conducted on novice players, and the authors themselves note that the link to problem gambling is argued rather than demonstrated in that study. These caveats are the reason the conclusions above are drawn narrowly.

Sources

Claim	Source
12.7% of gamblers, 48.5% of revenue (Australia, to March 2024)	Roy Morgan PGSI data via AGR
Revenue share from problem gamblers: 40.2% France, 31.6% Québec, 32% Germany	Fiedler, Kairouz et al., Journal of Business Research, 2019
Cross-country revenue concentration, 20+ studies	GREO evidence brief, 2019 (PDF)
Losses disguised as wins: 18.4% of spins at 15 lines; arousal equal to real wins	Dixon, Harrigan, Sandhu, Collins & Fugelsang, Addiction, 2010 (PDF)
Near-miss effects on reward circuitry	Murch & Clark, The Neuroscientist, 2016
Time on device, the machine zone, 15 years of fieldwork	Natasha Dow Schüll, Addiction by Design, Princeton University Press, 2012
Multiple RTP builds; operator selects by margin target	BGaming, provider explainer
£19.2m penalty; itemised customer failures; 331 self-excluded customers	Gambling Commission public statement, 28 March 2023
A$450m AUSTRAC penalty against Crown Resorts, 2023	ABC News
CSGO Lotto: concealed ownership, FTC action	FTC press release
Curaçao LOK reform, obligations, fees, tax treatment	Legal 500 Curaçao gambling law guide

Reproducing the maths. The expected-loss figures are turnover multiplied by house edge. The progression table multiplies the stake by 1/(m−1) of cumulative deficit at a 1.3026x target, and the streak probabilities are 0.24 raised to the number of consecutive losses. Any reader can rebuild both in a spreadsheet in under five minutes, and is encouraged to.