Why the feed you think is random is anything but, and why the code you think is rigged is legally required to be fair.
Two kinds of software shape most of what people see online. The first is designed to reduce randomness: it learns what you will engage with and serves more of it. The second is required by law to be statistically unpredictable and immune to any pattern analysis. Most people assume the opposite is true about both. What follows draws on the official record – regulatory filings, certification standards, and peer-reviewed cognitive science – to explain why.

AT A GLANCE
The core split: Predictive algorithms (engagement-tuned, deterministic) vs. certified RNGs (statistically unpredictable, regulator-audited)
US framework: Nevada Gaming Control Board (Regulation 14) · NJ Division of Gaming Enforcement (N.J. Admin. Code §†13:69E-1.28G)
Canadian framework: AGCO Registrar’s Standards for Internet Gaming (in force April†4, 2022) · iGaming Ontario
Global certification: Gaming Laboratories International (GLI-19, GLI-16) · ISO/IEC†17025 accredited
Documented cognitive error: Gambler’s Fallacy – described by Laplace (1796), formalised by Tversky & Kahneman
Start here: two types of code, two opposite goals
The digital environment presents users with two fundamentally different kinds of software, yet they are routinely confused with one another.
Predictive algorithms – social feeds, search rankings, trading-app notifications are built to be reactive and deterministic. The code’s job is to model what you will click next and surface it. There is no dice roll. When users feel their feed is “suppressing them randomly” the more accurate framing is: the scoring model is not working the way they assumed. The output feels unpredictable because the model is complex, not because it is random.
Certified RNGs – the software that determines outcomes in regulated iGaming and lottery systems are built to be the opposite: statistically unpredictable, immune to external influence, and provably unresponsive to prior outputs. This is not a marketing claim. It is a condition of the operating licence. Users who believe online games “tighten up” after a win, or are “due” for a payout, are attributing memory to a system specifically engineered and independently audited to have none.
The paper trail: how online casino RNGs are regulated in the US and Canada
The clearest place to see a certified RNG obligation in action is the regulated iGaming market. Here is what the regulators in three North American jurisdictions require.
Nevada (NGCB Regulation 14). The foundational document for US gaming RNG requirements. Regulation 14 mandates that games use an RNG “impervious to influences from outside the device” that passes statistical confidence testing. The NGCB maintains a public registry of approved testing laboratories, including Gaming Laboratories International (GLI).
New Jersey (N.J. Admin. Code §†13:69E-1.28G). The largest regulated online casino market in the US. The code specifies that RNGs must be “statistically independent, conform to random distribution and pass variously recognized statistical tests such as the Chi-square test, Mono-bit test, and the Runs test.” RNGs must also “employ a random seed that is determined by an uncontrolled event” – explicitly preventing predictable cycling.
Ontario (AGCO / iGaming Ontario). Since the regulated market opened on April 4, 2022, operators must certify their RNGs to the AGCO Registrar’s Standards by an independent testing laboratory registered with the AGCO before any system goes live. The iGaming Ontario Act, 2024 (in force May 12, 2025) restructured iGaming Ontario as a fully independent Crown agency.
A GLI-19 or GLI-16 certificate is the primary public document confirming a game’s RNG was independently evaluated. GLI operates as an ISO/IEC†17025 accredited testing laboratory.
Cognitive errors: why we mistake noise for a pattern
The Gambler’s Fallacy (Monte Carlo Fallacy). A player sees the same outcome repeat eight times and concludes the opposite is “due.” A certified RNG has no memory. The probability of any outcome on the next spin is identical to what it was on the first. Tversky and Kahneman traced this error to the representativeness heuristic: people expect small samples to mirror the distribution of a much larger population. Clotfelter and Cook (1991) documented the same pattern in Maryland lottery data, wagers on a number dropped sharply after it was drawn, then gradually recovered, a measurable imprint of a belief with no mathematical basis.
The Clustering Illusion. True randomness produces streaks and dry spells. A perfectly random sequence of one million coin flips contains long runs of heads that look, to the human eye, like a malfunction. When platforms display “Recent Big Wins,” they surface a statistically inevitable cluster from a large dataset – not evidence that a game is “hot.” The same shortcut makes a repeated headline feel like corroboration. Repetition is an algorithmic output, not a verification of fact.

The algorithmic feed exploits this same wiring
A persistent misconception frames social feeds as “slot machines” – random, keeping users in suspense. The analogy inverts the reality. A certified RNG is required to be unpredictable. A social feed is engineered to be the opposite: to predict your next action and maximise dwell time.
Meta’s transparency documentation describes ranking as based on “features of the content, and attributes of individuals and their interactions” – a scoring model, not a dice roll. When users believe their feed is random, they cannot understand why they see what they see. That gap is exploitable by misinformation actors who claim the algorithm is suppressing truth. The mundane reality: it optimises for engagement, and high-emotion content scores well.
Common questions
Yes. N.J. Admin. Code §†13:69E-1.28G and the AGCO Registrar’s Standards both mandate that RNGs pass statistical independence tests and be certified by an accredited lab before deployment. Operating a non-compliant game carries licence revocation.
Because RNG fairness and house edge are separate concepts. The RNG ensures each outcome is statistically independent. The house edge, disclosed in every game’s rules (e.g., 2.7% in European Roulette, 4–10% in slots), ensures the operator is profitable over a large volume of outcomes. They are independently audited under different rule provisions.
No. Per both companies’ own documentation, feeds rank content on a scoring model based on predicted engagement, recency, and relationship signals. A certified RNG is designed so past outputs cannot influence future ones. An engagement algorithm is designed so past behaviour entirely determines what comes next. They are structural opposites.
A NOTE ON SOURCING
Regulatory citations: N.J. Admin. Code §†13:69E-1.28G · AGCO Registrar’s Standards. Cognitive science: Tversky & Kahneman (1971); Clotfelter & Cook (1991). Platform transparency: Meta AI system cards · X open-source ranking repository (GitHub).
This page is informational only and does not constitute gambling, financial, or legal advice. Problem gambling support: National Problem Gambling Helpline 1-800-522-4700.
