Slot RNG: how it is tested, and what a certificate leaves open
Every spin result comes out of a random number generator, and it's picked the moment you hit spin, not when the reels stop. Labs like GLI check that generator before a game can go live: they read the code and push millions of outputs through statistical tests. Which RTP version your casino switched on is a different matter, and you find that in the game's rules screen.
What a lab has to check
| Clause | What it demands |
|---|---|
| 3.2.1 Source code review | The lab reads the RNG, scaling and shuffling code and looks for bias, errors, hidden switches |
| 3.2.2 Statistical analysis | Tests on the final outcomes, judged together at a 99% confidence level |
| 3.2.2 Named tests | Chi-square, overlaps, coupon collector, runs, interplay correlation, serial correlation, duplicates |
| 3.3.1 Strength | Cryptographically strong: no prediction even with the source code in hand |
| 3.3.2 Seeding | Not seeded from a time value alone; no two games start from the same seed |
| 4.5.2 No near-miss swap | If the RNG picks a loss, the game may not show a different, closer-looking loss |
| 4.5.2 No memory | Bonus chance and RTP may not adjust to past payouts |
| 4.7.1 Minimum return | At least 75% RTP unless the regulator sets another figure |
| 4.7.3 Top prize odds | At least 1 in 100,000,000, or the real odds must be shown |
GLI-19 version 3.0, revised 17 July 2020. Regulators adopt it in full or in part; a v4.0 draft was out for comment in July 2026.
In Britain the Gambling Commission wants results that are 'acceptably random', provable with standard tests, and it bans compensated games. That's the idea of a machine that tightens up after paying out and loosens after a dry spell, which is how a lot of players still think slots work. People swear a game owes them after 300 dead spins. Under these rules the next spin has the same odds as the first one of the night.
Could you spot a wrong RTP yourself?
| Standard deviation per spin, in x the bet | Within 2 points | Within 1 point |
|---|---|---|
| 5 | 240,100 | 960,400 |
| 10 | 960,400 | 3.8 million |
| 34.7 (median, 100 most played Engine games) | 12 million | 46 million |
| 50 | 24 million | 96 million |
| 100 | 96 million | 384 million |
Our calculation: spins = (1.96 x standard deviation / margin) squared, 95% confidence. Median standard deviation from Engine Insights. Bonus buys swing far more per bet, so they need even more.
A typical Engine slot swings about 35x the bet per spin, going by the median standard deviation of the most played ones. After 10,000 spins your result can still be tens of points off the RTP, and separating a 96% game from a 94% one takes something like 12 million spins. Nobody plays one game that long. Skip the session spreadsheet, open the rules screen and compare the RTP line with the studio's list. Hacksaw prints four settings per game, and Le Bandit runs from 96.34% down to 88.36%.
| Game | Studio | Wagered | Bets | Configured RTP | Paid back |
|---|---|---|---|---|---|
| Gates of Olympus Super Scatter | Pragmatic Play | $398.0m | 61.3m | 96.50% | 96.44% |
| Wanted Dead or a Wild | Hacksaw Gaming | $158.8m | 21.8m | 96.38% | 95.47% |
| Sweet Bonanza 1000 | Pragmatic Play | $110.2m | 29.7m | 96.53% | 94.22% |
| Buried By Night | Hacksaw Gaming | $60.0m | 15.4m | 96.32% | 105.34% |
| Wanted Salvation | Hacksaw Gaming | $76.1m | 8.5m | 96.38% | 79.84% |
Paid back = 1 minus GGR / wagered in the window (bonushunt.gg provider analytics). A bonus buy counts as one bet at its full price. bonushunt.gg notes its data can have gaps or lag.
The busiest game in the table landed close to its setting. Smaller ones moved a lot further, in both directions, over tens of millions of dollars. A few max-win hits or buys at big stakes are enough to do that inside a month. The RTP page has the same comparison for the games with the most bets.
The labs
| Lab | What it publishes | Detail |
|---|---|---|
| GLI (Gaming Laboratories International) | Its standards, GLI-19 for online games, as free PDFs | Named by many regulators as the technical standard |
| eCOGRA | RNG testing for operators and studios | Methodology approved in Great Britain, the Netherlands and Spain, among others |
| BMM Testlabs | RNG and game maths evaluation | Certified Chainlink VRF, an on-chain random number service, in September 2022 |
A certificate is usually linked in the casino footer or the game's info screen, and it names the game and the version. If the name or version on it differs from what's on your screen, it was issued for some other build.
Stake Engine works differently. Each game mode ships as a fixed table of outcomes with weights, the RNG picks one entry per round, and the tables are the maths the studio submitted. How to replay a result from a seed pair is on the provably fair slots page.
Not affiliated with GLI, eCOGRA, BMM Testlabs, Pragmatic Play, Hacksaw Gaming or bonushunt.gg. Names belong to their owners. This page is not an assessment of any game. 18+, play responsibly.
Questions
Are online slots rigged?
Certified slots use an RNG that a lab has tested, and standards like GLI-19 forbid games that adjust to past payouts. What a casino can change legally is the RTP setting, so read it in the rules screen.
How do labs test a slot RNG?
They review the source code and run statistical tests on the final outcomes, for example chi-square, runs and serial correlation, judged at a 99% confidence level under GLI-19.
Does a slot get colder after a big win?
Not in a certified game. Each result is independent, and GLI-19 clause 4.5.2 forbids changing the bonus chance or the return based on earlier payouts.
How many spins does it take to check a slot RTP?
Millions. At the median Engine standard deviation of 34.7x the bet, pinning the RTP within 1 point takes about 46 million spins.
Sources
- GLI-19 Standards for Interactive Gaming Systems v3.0 (PDF). sections 3.2, 3.3, 4.5.2, 4.7.1, 4.7.3; revised 17 Jul 2020 Checked 12 November 2026
- UK Gambling Commission: RTS 7, generation of random outcomes. 'acceptably random', compensated games not permitted Checked 12 November 2026
- eCOGRA: RNG testing. methodology approved in Great Britain, the Netherlands and Spain Checked 12 November 2026
- European Gaming: BMM Testlabs certifies Chainlink VRF. 2 Sep 2022 Checked 12 November 2026
- Engine docs: How the RGS works. books, lookup tables, weighted random selection Checked 12 November 2026
- Hacksaw Gaming: Le Bandit. RTP settings 96.34% / 94.23% / 92.17% / 88.36% Checked 12 November 2026
- Engine Insights (public API, snapshot). 30 days, 13 Oct to 11 Nov 2026 Checked 12 November 2026
- bonushunt.gg provider analytics (Stake slot statistics). month and all-time windows Checked 12 November 2026
18+ · Play responsibly · gamblingtherapy.org