A sound RNG can still feed a biased game if random values are mapped incorrectly. Mapping and payouts need separate checks.
From random value to result
Software generates random values, which the game maps to cards, symbols or other events. For operators under the Online Casino Gambling Act, New Zealand’s 2026 standards address both stages. Software pseudo-random generators must be cryptographically secure, with unpredictable and properly distributed output. Scaling and mapping must preserve the certified probability distribution. A predictable starting state can undermine a sound algorithm. The standards require initialization and reseeding to avoid predictability and prohibit two instances from synchronising to the same stream. Operators must continuously monitor output integrity and deactivate an affected game when they detect an anomaly or failure.
The standards prohibit adaptive or compensated behaviour and secondary manipulation of selected outcomes. A game cannot offset a previous win or loss by changing the probability of the next unrelated chance event. Near-miss animation does not alter this requirement.
Eight outcomes: hit rate and RTP are not the same number
Consider an invented game with a one-unit stake and eight equally likely values, 0–7. Values 0–3 pay nothing, 4–6 return the stake and 7 returns four times the stake.
| Random values | Probability | Total return per one-unit stake | Contribution to expected return |
|---|---|---|---|
| 0–3 | 4/8 | 0 | 0 |
| 4–6 | 3/8 | 1 | 3/8 |
| 7 | 1/8 | 4 | 4/8 |
Expected return is 7/8, or 87.5% RTP. Four of eight outcomes pay something, giving a 50% hit rate, but only one returns more than the stake. If draws are independent, a losing streak changes none of these probabilities and neither percentage predicts the next result.
Now suppose the generator emits ten equally likely values, 0–9, and the software takes the remainder after division by eight. Positions 0 and 1 each receive two inputs; the other six receive one each. Both overweighted positions lose in this invented paytable, so expected return falls from 7/8 (87.5%) to 7/10 (70%), and hit rate from 4/8 (50%) to 4/10 (40%). The mapping is biased despite the uniform input. New Zealand’s standards require scaling and mapping to preserve the certified distribution.
For the ten-value example, one simple unbiased mapping would reject values 8 and 9 and use 0–7 directly; each accepted position then has one input. That method discards one fifth of draws and would need an implementation that handles retries correctly. A larger uniform source can use an equivalent rejection boundary. The point is to test the actual mapping, since a mathematically sound generator can still feed a biased game.
RTP, hit rate and variance
RTP derives from the probability and payout of every outcome. Hit rate counts defined awards; variance describes the spread of results; the maximum award caps the return under the rules. None predicts a single session, and even a high-RTP game can produce a complete loss for a player.
The RTP shown on a platform must describe its active variant, even when a supplier offers several configurations. New Zealand's standards require accessible rules, winning odds and theoretical RTP expressed as a percentage. If equivalent free and paid versions differ in odds or return, the operator must explain the difference before a stake.
A jackpot may receive contributions, a bonus purchase can increase the round cost and a feature can have its own probability. The paytable needs to be read with eligibility and stake conditions for the available variant.
Testing scope
Statistical tests can detect departures from an expected distribution if the sample and hypothesis are adequate. Code review can find discarded or altered random values. Integration tests check whether the accepted wager settles once and the displayed result matches the ledger.
Gaming Laboratories International publishes interactive gaming standards covering random outcomes and game selection. A certificate needs a named component, version and jurisdiction. Neither the laboratory's name nor its standards establish an operator's legal status or that a deployed game was assessed under New Zealand's rules.
Physical live games need evidence of equipment quality, controlled access, surveillance and dealing procedures. New Zealand's standards address live dealers separately from RNG. Where a brand has both formats, its round record must identify whether software or a physical event produced the result.
A useful test report names the generator, outcome mapping, paytable and exact build. GLI-19 v4.0 examines RNG and game outcomes separately: a pass on the raw number stream cannot establish the distribution after mapping or whether the rule sheet matches the active configuration.
A “provably fair” check reproduces a disclosed computation from committed inputs. GLI-19 v4.0 sections 5.5 and 6.6.3 cover the commitment, later seed reveal, client input, round identifier where needed and published algorithm. A matching calculation does not establish the game’s RTP or whether the account ledger settled it correctly.
Reproducing BGaming’s published roulette example
BGaming documents a SHA-256 commitment to Result + Secret before the bet, followed by disclosure after the round. Its American Roulette example uses this exact UTF-8 input, with no spaces or trailing newline:
{"server":2,"game":"AmericanRoulette","secret":"290fb3c9fbfc2afa5339bf27e5c25cbb"}
The SHA-256 digest is:
f88e5c0943740b0269b21347fa26c59b1717ee200584ac40e1c56edd49ad82b8
That matches BGaming’s published hash. The next step uses the game’s disclosed conversion: its example combines the server value 2 with Client Seed 5 to produce Final Result 7. Matching the hash checks the revealed input; applying the conversion checks the stated result. Extra spaces produce a different digest, so the exact input bytes matter. Other games can use different input formats and algorithms.
Passing a randomness test leaves another question
“statistical testing cannot serve as a substitute for cryptanalysis”
NIST SP 800-22 Rev. 1a
A sample can pass distribution tests while leaving seed secrecy, internal state and implementation weaknesses unexamined. An assessment therefore needs the generator design and its security evaluation alongside the statistical results.
NIST's document concerns statistical testing of random and pseudorandom sequences for cryptographic applications. It is not a casino approval, a paytable assessment or evidence that a particular game has been tested.
Player-facing mechanics and failures
The rules should identify when a round starts, which actions affect it and when the result becomes final. New Zealand's standards require a separate deliberate start, at least 2.5 seconds from one round's start to the earliest next start, and time tracking. Winning positions in predetermined-pattern games must not change except as the rules allow.
After an interruption, the platform needs enough records to restore events where possible. If the game cannot continue, the standards require winnings reached or returned bets, whichever is better for the customer. A controlled connection test can record one recovery; a screenshot alone cannot prove the implementation.
DIA’s account guidance requires identity and age checks before activation. If chance-game play becomes difficult to control, Safer Gambling Aotearoa offers free support and a 24/7 helpline on 0800 654 655.