Random number generators: how a slot picks a result, and who tests it
In a slot tested to GLI-19, every chance event that affects a result must come from a random number generator, or RNG (clause 6.6.2). GLI-19 is a standard written by Gaming Laboratories International, and any regulator may adopt it in whole or in part under its own policy (1.2.1(d), 1.2.3). Version 4.0 is dated 3 November 2026; we found nothing showing which regulators have moved to it, so its new rules are marked. Britain’s Gambling Commission publishes its remote technical standards (RTS) and testing strategy in full, so Britain is our worked example. GLI-19 and the British rules are written for money games, and neither is South African law; we found no South African text adopting either. Every demo here runs on virtual credits. Whether slots are rigged is answered near the end, from these rules.
What is a random number generator in a slot?
A slot RNG produces the numbers behind a game’s results. GLI-19 describes it as a device, algorithm or system designed to give values that cannot be told apart from random selection; version 4.0’s definition adds the word unpredictable (glossary). The Commission’s definition is shorter: hardware or software that generates numbers meant to be statistically random (definition of terms).
GLI-19 names three kinds (5.1.1): software, working from an algorithm; hardware, drawing on small physical events such as thermal noise; and mechanical, such as wheels and shufflers. GLI-19 asks for output that cannot be predicted (5.3.1), and the British rules ask for results shown to be random to a high degree of confidence (RTS 7A). None of the texts we read requires a physical source of randomness, since software generators are allowed, and no studio page we read says which kind its slots use.
How does a number become the symbols on the reels?
The raw number is scaled to the range the game needs, then mapped to a result, such as the symbol at a stop on a reel strip (definition of terms; GLI-19 glossary). The Commission’s example scales 1 to 100,000 down to 1 to 52 for a pack of cards, and Figure 1 does the same for an invented reel.
GLI-19 asks that every possible number be equally likely and that scaling and mapping be unbiased, allowing values to be discarded where that removes bias (5.2.3). In the figure, 100,000 divides evenly into 10 stops; an uneven split would favour some stops, on our reading the kind of bias the rule forbids.
Equal numbers do not mean equal symbols. Where the design sets an uneven spread, the final result must follow it (5.2.3). Version 4.0 adds that mapped results must match the approved paytable or reel strips and follow any weights without bias (6.6.2(c)). Weights are set chances that make some results more likely. As we understand it, the symbols on screen do not show how often each is set to land.
A spin can need more than one draw. Habanero’s page for Lucky Lucky, a 3x1 game with one line, says: “Wild reveals x2 or x3 multiplier at random.” We take that to be another random pick within the spin; Habanero does not say how it is drawn. Where one draw produces several values, GLI-19 asks that knowing some of them tell you nothing about the others unless the design provides for it (5.2.4), and that each generator, and each way a game uses one, be evaluated separately (6.6.1(b), (c)).
Is the result settled before the reels stop?
On our reading, yes. GLI-19 asks that events of chance affecting a result come only from an approved RNG (6.6.2), and version 4.0 adds that results be used as soon as they are drawn, in order, as the design provides (6.6.2(d)).
Both GLI-19 and the British rules bar a game from swapping in a different losing result to show the player (6.6.2(e); RTS 7C), and Britain bars any design that could reasonably be expected to mislead about how likely results are. New in version 4.0, GLI-19 uses “near-miss” for a presentation that suggests a better chance at an advertised top award or bonus than the maths gives (glossary), and bars showing these more often than the maths would produce them (6.7.1(a)(iii)). No text we read says where a losing result that looks close becomes misleading, and the pay system changes what looks close.
In Britain a casino game may not let the player cut the time until the result is shown (RTS 14E). The Commission’s guidance gives turbo, quick spin and slam stop as examples and exempts features played without an extra stake. As we read the rule, a stop only shows the result sooner. Timing a press cannot steer the draw either, on our reading, because GLI-19 asks that the generator stay unpredictable (5.3.1, 5.3.2(a); see what a test lab checks). A cascade or a feature may still draw again as the spin goes on.
Can one spin tell you anything about the next?
Not about the next draw, where GLI-19 applies. It asks that one draw give no information about later ones (5.2.4) and that, except as a game’s rules provide, chance events be independent of earlier games (6.7.1(c)). It also asks that every result stay possible on every draw, whatever came before, unless the design says otherwise (5.2.5). A note new in version 4.0 adds that a rare event can then happen again soon.
A game’s rules can carry something forward, like Sugar Rush 1000’s marked spots in free spins, without making the next draw predictable. NetEnt says the jackpot in Mega Joker is “awarded on a randomly selected game round” and that “The Jackpot value increases each time the basic game is played.” Growing is not the same as due; NetEnt does not say whether the chance changes as it grows. Under GLI-19, a run of losses alone says nothing about the next draw; the volatility guide shows how often long dry runs turn up.
Can a game change its odds after it pays out?
Not in Britain, which bans compensated games, those that adjust the chance of winning to past payouts or takings; the Commission also calls this adaptive behaviour (RTS 7A; definition of terms).
GLI-19 does not ban it outright. A game may not discard or change a drawn result because of adaptive behaviour (6.6.2(b)). A return to player (RTP) that follows past payouts, or a bonus chance that depends on earlier awards, is allowed only where the game’s rules provide for it (6.8.1(b), 6.7.1(c)). Version 4.0 adds that the bonus case is allowed only if the player is told (6.9.1(c)).
In Britain a game’s rules, payouts and odds may not change while it is open for gambling, except as its rules provide, and players must be told (RTS 7D). The licensee tests this, and no outside lab is required (testing strategy, Table 1). An operator’s choice among a game’s RTP settings is a separate matter (the RTP guide).
What does a test lab check?
The code and what it produces, where a regulator requires testing to GLI-19. The lab reads the source code for the randomness, scaling, shuffling and anything else critical to the result, looking for bias, errors, malicious code and undisclosed switches (5.2.1).
Statistical tests on the final results, after scaling and mapping, check the intended spread and the independence of draws. GLI-19 lists seven tests a lab may use, including chi-square, runs and serial correlation; those applied are judged together at a 99% confidence level (5.2.2). We read that as a statistical threshold for the whole set, not a fairness score. No sample size is set, only enough data to detect significant deviations.
GLI-19 asks that the generator stay unpredictable even to someone who knows its source code and past outputs (5.3.1, 5.3.2(a)). Its seed, the value that sets its starting state, may not come from the clock alone (5.3.3(a); glossary). The game’s maths is checked separately: in Britain an approved test house examines the game through maths verification, source code and play (testing strategy, Table 1).
Is the copy a casino runs the version that was tested?
In Britain it is supposed to be. Before release, an approved test house, which the licensee appoints from the Commission’s list, checks the generator and the game against RTS 7A, 7B, 7C and 7E (testing strategy, Table 1; Approved test houses). The game may go live only once that report is with the Commission, and the report should name the build by its software number and digital signature (procedure for testing).
Meeting these standards is a licence condition (RTS introduction, 1.3). NetEnt says its RNGs are “tested and certified by independent accredited test facilities (ATF)”, which check its products against the regulations “for the regulated markets in which we operate”. As we understand it, a certificate covers one build in one market.
A change that may affect fairness, through the generator, the scaling and mapping or the game rules, is a major update needing retesting by an approved test house and an updated report with new digital signatures. Minor ones, such as a sound-format fix, do not need outside retesting (Annex A; procedure for testing). On our reading, certification does not change the RTP or the house edge, or make a game good value. Copies of real games also circulate; Play’n GO says some providers offer copies of its games without the rights to them, and warns: “Worst case, they may not be properly certified and verified for fairness.”
What do South Africa’s testing rules cover?
The texts we read cover gambling machines and devices. The National Gambling Board (NGB) says these must meet SANS 1718 and be tested by gambling laboratories with a regulator’s licence and SANAS accreditation to ISO 17025. Under the National Gambling Act as published in 2004, a testing agent is licensed by a provincial licensing authority, SANAS-accredited, and independent of other licensees, any regulator and the South African Bureau of Standards (SABS) (s 24). It tests against the applicable standard and reports to the requester, the province, the Board and the SABS (s 25).
The Act’s gambling machine definition includes software; we could not establish whether it reaches online games. SANS 1718, sold by the SABS, has part titles such as casino equipment, limited payout machines, jackpot equipment, roulette wheels and server-based gaming systems, the last described for play on licensed premises. None of the entries mentions online games or RNGs, and we have not read the standard. The NGB’s Testing (Testlabs) page lists no laboratories; provincial boards were not checked.
Online, the NGB’s FAQ answers “Is online gambling legal?” with “No”, quoting section 11: “A person must not engage in or make available an interactive game except as authorised in terms of this Act or any other national law.” We found no South African standard, rule or lab list for online slots or their generators.
So, are online slots rigged?
On a site a regulator oversees, the rules ban rigging, but only in the ways that regulator’s rules name. British rules forbid compensated games and designs that could reasonably be expected to mislead about the odds (RTS 7A, 7C), and let odds change in play only as the game’s rules say (RTS 7D). Where GLI-19 applies, a game may not change a drawn result because of adaptive behaviour or swap one losing result for another (6.6.2(b), (e)). An RTP or bonus chance may follow past results only where the rules provide for it (6.7.1(c), 6.8.1(b)).
What stays legal still costs the player: the house edge in the maths, with RTP display left to each regulator (6.8.2), and an operator’s choice of a lower-RTP version (the RTP guide).
Testing can miss things, so the British rules add checks after release. Licensees must monitor actual RTP for under- and overpayments (testing strategy, Live RTP monitoring). Those that have taken on responsibility for games testing generally also face a yearly audit by an approved test house, which checks a sample of their updates and the list of games they offer (Annual games testing). The testing strategy adds that telling a player who complains about fairness that the game met the testing standards is not enough, because errors can get past testing. These are British rules. For copied games, see Play’n GO’s warning above.
Nothing we read allows a player to confirm which certified version a site runs, and this site cannot check either; we vouch for no casino or game. For South Africa, see section 11 of the Act, quoted above.
Does a demo on this site use a tested generator?
We do not know. Under GLI-19, a game’s free-play mode should represent the paid game’s normal operation and not mislead about the chance of winning (6.11.1(a)). In Britain a play-for-free version on the same website as the money game must use the same rules and, as far as the operator reasonably can, reflect its chance of winning (RTS 6A). The Commission’s guidance says it should use the same generator, another that meets RTS 7A or a public one expected to show no systematic bias. The guidance adds that a supplier’s own site showing a game that pays more than normal should say it is a demonstration of the bonus features. On our reading, then, a demo away from a casino’s site may not match the paid game.
We do not configure the demos and cannot see which maths version or generator any of them runs. They run on virtual credits: nothing is deposited, nothing is paid out, and a demo is not practice for playing with money.
If gambling has stopped being fun for you or someone close to you, the responsible gambling page lists where to find help in South Africa.
Sources
Pages were read on 4 November 2026; the studio, NGB FAQ and gov.za quotes were matched to their pages again on 5 November 2026. GLI-19 and Commission text is paraphrased with clause numbers.
- Gaming Laboratories International, GLI-19 Standards for Interactive Gaming Systems, version 4.0, dated 3 November 2026, the version linked on the GLI standards page: sections 1.2.1, 1.2.3, 5.1.1, 5.2.1 to 5.2.5, 5.3.1 to 5.3.3, 6.6.1, 6.6.2, 6.7.1, 6.8.1, 6.8.2, 6.9.1 and 6.11.1, and the glossary. The paytable and cascades guides cite version 3.0, revised 17 July 2020; the table below maps its clause numbers to version 4.0.
- UK Gambling Commission, Remote gambling and software technical standards: the introduction, definition of terms, RTS 6, result determination for play-for-free games, RTS 7, generation of random outcomes and RTS 14, responsible product design.
- UK Gambling Commission, Testing strategy for compliance with remote gambling and software technical standards: Approach (Table 1), Procedure for testing, Annual games testing, Live RTP monitoring and Annex A, major and minor game and software updates.
- UK Gambling Commission, Approved test houses.
- National Gambling Board: Technical Standards (SABS), Testing (Testlabs) and FAQs.
- National Gambling Act 7 of 2004, Government Gazette No. 26670, 12 August 2004, in the NGB’s copy: the definitions of “gambling machine”, “interactive game” and “testing agent”, and sections 11, 24 and 25.
- National Gambling Amendment Act 10 of 2008, on gov.za, which describes it as an Act “to provide for the regulation of interactive gambling” and lists its commencement as “To be proclaimed”.
- SABS webstore entries for SANS 1718 parts 1, 2, 3, 4, 5, 8, 9 and 10.
- NetEnt: Responsible Gaming and the Mega Joker game page.
- Play’n GO: Licences and Certification.
- Habanero: the Lucky Lucky game page.
| Version 3.0 | Version 4.0 | Note |
|---|---|---|
| 1.3.1(d), (g), 1.3.3 | 1.2.1(d), (g), 1.2.3 | |
| 3.1.1 | 5.1.1 | 4.0 adds combinations of the three kinds |
| 3.2.1, 3.2.3, 3.2.4 | 5.2.1, 5.2.3, 5.2.4 | 4.0’s 5.2.3(a) also allows delaying or changing values to remove bias, not only discarding them |
| 3.2.2 | 5.2.2(a) to (c) | 3.0: one paragraph, then the seven tests as (a) to (g); 4.0 splits the paragraph into an opening, (a) and (b), and lists the tests under (c); the 99% level is in both |
| 3.2.5 | 5.2.5 | 4.0 adds the internal state and a note on rare events |
| 3.3.1 | 5.3.1 | 4.0 adds the design and allows a comparable level of unpredictability |
| 3.3.2(a) | 5.3.2(a) | |
| 3.3.2(b) | 5.3.3(a), in part | the time-value rule is 5.3.3(a); the rest of 3.0’s (b) is now 5.3.3’s opening, (b) and (d); 4.0’s 5.3.2(b) uses the name Known Input Attack for a different attack |
| 4.3.3(a), (d), (e) | 6.3.5 opening and (a), (g), (h) | the transfer of awards in 3.0’s (d) is now 6.3.5(f) |
| 4.3.4(e) | 6.3.2(e) | |
| 4.4.1(a), (b), (d), (e) | 6.4.1(a), (b), (e), (f) | |
| 4.4.1(g) | 6.4.1(h) | 4.0 adds detail on multiplier rules |
| 4.4.1(j) | 6.4.1(k), 6.5.2(d)(iii) and (v) | similar, wording differs; the per-line and line-plus-scatter cases moved to 6.5.2 |
| 4.4.1(k), (r) | 6.4.1(l)(i), (o) | |
| 4.4.1(l), (m), (n) | 6.4.2; (m) and (n) together in 6.4.2(f) | symbols and objects now have their own section |
| 4.4.3(e) | 6.5.2(d) | line games are now covered under reel-based games |
| 4.5.1(a), (b) | 6.6.1(b), (c) | |
| 4.5.2 opening | 6.6.2 opening | similar, wording differs |
| 4.5.2(b) | 6.6.2(b) | 4.0 adds delay, substitution and changes based on value; (b)’s second sentence, use outcomes as the rules direct, is now in 6.6.2’s opening |
| 4.5.2(c) | 6.6.2(e) | 4.0 adds “or similar outcome presentation” |
| 4.5.2(d) | 6.7.1(c) | |
| 4.5.2(d)(i) | 6.9.1(c) | similar, wording differs: 3.0 has no disclosure wording, only (d)’s exception for the rules of the game |
| 4.5.2(d)(ii) | 6.8.1(b) | |
| none | 6.6.2(c), 6.6.2(d), 6.7.1(a)(iii) | new in 4.0; 3.0’s 4.5.2(b) says only that outcomes be used as the rules direct |
| 4.7.2 | 6.8.2 | RTP display is now RTP disclosure |
| 4.7.2(c) | 6.8.2(c), in part | 6.8.2(c) covers progressive and incrementing bonus jackpots only; for 3.0’s limited time awards and other bonuses or features we found no clear counterpart |
| 4.9.1(a), (b) | 6.11.1(a), (b) | 4.0 adds the word “demo”; (b) now says clearly identified, not prominently displayed |
| Glossary: Artwork, Game Cycle, Mapping Algorithm, Paytable, RNG, Scaling Algorithm, Seeding/Seed | the same entries | RNG: 4.0 adds the word “unpredictable” |
| Glossary: Near Miss, a top-award combination shown above or below an active payline | Near-Miss | new sense in 4.0: a misleading design |
