Stored per round
The model and its fingerprint, the bank's fingerprint, the sealed secret, its published fingerprint, the player's number, the position in the sequence, what was asked for, where the reels stopped and the full result.
Fairness
A player has to take a game's word for a great many things. This page is about the one thing they do not have to take on trust, and about what we would hand a testing laboratory that came to check the rest.
Before the round, a secret number goes onto a card and the card is shut. Nobody has seen it, including the player and including the operator.
A short string that could only have come from this card, and from which nobody can work backwards to what is written on it. It goes out while the card is still sealed.
Now, and only now. The outcome is worked out from both numbers together, so neither side alone decides what the board does.
Afterwards the card comes out. The player, their operator, or a regulator can check it against the fingerprint published beforehand, put both numbers back through the same mathematics and arrive at the same board, the same lines and the same payout.
We could not have swapped the card after seeing the player's number, because the fingerprint was public before they chose it. The player could not have steered the result, because the card stayed shut until the round was over.
That is the whole argument, and it is the only part of a slot machine that can be argued without taking anyone's word for anything.
Most of the argument about a slot machine is really an argument about what it is allowed to produce. Our games do not improvise. Every board they can ever show was built in advance, given a weight, and evaluated for what it pays.
The bank is checked byte for byte when the service starts, and every board is re-evaluated against the rules at that moment and must pay exactly what the bank says it pays. One altered byte and the service refuses to load rather than run on mathematics nobody approved.
The model and its fingerprint, the bank's fingerprint, the sealed secret, its published fingerprint, the player's number, the position in the sequence, what was asked for, where the reels stopped and the full result.
Feed those values back into the same model and the round reproduces exactly. Our own tests prove the current engine reproduces rounds recorded by the previous one, seed for seed, which is the property that lets a laboratory check history rather than only the present.
Money movements, configuration changes and administrative actions are journals rather than editable records. The database declines to rewrite them, so an audit is reading rather than reconstructing.
Every change to a title's settings carries the person who made it, the time, the reason and the difference. Turning real money on for a brand takes two people.
We are preparing for a GLI-19 style review of the random number generator and the return to player, and we have not yet submitted to a jurisdiction. We do not hold a licence today and we will not imply otherwise.
The pack is assembled rather than promised. It holds the generator and how it is seeded; the draw sequence and how a value in a range is taken from it without bias; the fixed bank, its fingerprint and the optimiser output it came from; the per-model return figures including every capped variant; the statistical test results; and the tool that turns a stored round back into its outcome.
The operating system's cryptographic generator, thirty-two bytes at a time. It is encrypted before it is stored and revealed only once the round it belongs to has finished.
No. A session records the configuration version it opened on and stays on it. A change made while somebody is playing applies to their next session, never to the spin in front of them.
It has its own bank, derived from the same boards, and it is built so that the expected payout of a purchase equals its price exactly. The mystery package picks between three tree counts with fixed weights before the first free round is drawn, and holds that choice for all five.
Values in a range are taken by rejection sampling over the full space rather than by a remainder, so every value in the range is equally likely. A long run of rejections raises an error instead of quietly returning a skewed value.
If you are reviewing us on behalf of an operator or a regulator, ask for the technical pack directly.