Order ofoperations.
Four disciplines, run in a fixed sequence, with the least popular one first. What follows is the order the work is done in, the rules that hold that order in place, and the argument for shipping fewer titles than we could.
- 01Mathematics
- 02Art direction
- 03Engineering
- 04Compliance
Four disciplines, run in one direction.
The order is not a preference. Each discipline hands the next one a set of facts instead of a set of options, and the sequence is chosen so that nobody downstream has to guess what an earlier decision will be. Swap any two of these steps and the game gets built twice.
Mathematics
Runs first, before a single frame exists.
The model is written as code: symbol set, reel construction, tumble behaviour, trigger conditions and the exact shape of the tail. It is simulated across hundreds of millions of rounds and optimised against the constraints a certifier actually measures — return, hit frequency, the distribution of the top percentile, and whether the maximum win is reachable in every mode rather than in one flattering configuration.
Every number this studio publishes is fixed here and nowhere else. Bassline Mars returns 96.50% and caps at 25,000× because the model does, not because either figure read well on a sell sheet. Settling it this early is uncomfortable — it means arguing about tail risk long before anyone has seen the game move — but the alternative is a feature quietly re-costed three weeks before submission, and that always shows on screen.
Art direction
Runs second, with the constraints already known.
Nothing is drawn to a game-sized canvas. Assets are modelled, lit and rendered at print resolution — masters run to 6,320 pixels on the long edge — and only then reduced to the tiers a browser downloads. The fidelity is spent rather than saved: what ships is a fraction of the source, but the detail was there to lose.
Working after the maths means art direction inherits facts instead of negotiating them. How many symbols there are, how often each appears, which set has to survive a twelve-spin round without turning into wallpaper — all of it arrives settled. That constraint is why the sets are specific. A brief with nothing fixed in it produces decoration; a brief with the mathematics already closed produces a place.
Engineering
Runs third, and is not permitted to decide anything.
One WebGL build serves portrait and landscape at sixty frames, with skeletal animation, streamed audio and art delivered in the same encoded tiers this site uses. The structural decision is the one that matters: the round is resolved before the first frame of the win plays. Presentation reads a result. It never produces one.
That separation is what makes a dropped frame, a skipped animation or a tab closed mid-tumble irrelevant to what was paid — and it is what makes the game auditable at all. The outcome is a record. Everything on screen is a rendering of that record, and can be rebuilt, replaced or switched off without touching it.
Compliance
Fourth on the list. Present from the first commit.
Restricted-word routing, jurisdictional copy variants and social-casino labelling are built into the game rather than bolted on at submission. Every player-facing string passes through a locale layer that knows which markets forbid which words, so opening a new jurisdiction is a configuration change and not a rebuild.
The certificate is treated as a design input, sitting alongside return and volatility from the first day of the model. Studios that treat certification as a final exam end up maintaining two versions of the same game: the one the regulator saw, and the one players get. There is one build here, and the submission documentation describes it exactly.
Anything left until the end gets decided under a deadline. Deadlines have a look, and players can see it.
That is the entire case for the sequence. It is not that mathematics matters more than art. It is that a constraint discovered early is a brief, and the same constraint discovered late is a compromise — usually one made at speed, by whoever happens to be closest to it. Every rule below exists to move a decision earlier.
Some work exists to settle an argument.
This render has never appeared in a spin. It was made to answer one question — how big is a stack, really, standing next to a person — and it answered it well enough that every set built afterwards inherited the same sense of size. It is the reason the base game reads as a room rather than a backdrop.
Work like this is difficult to justify on a schedule and impossible to fake afterwards. A world either has a consistent scale or it does not, and a player notices the difference long before being able to name it.
A standard is only real when it costs something.
These are the working rules. Each one has, at least once, been the more expensive option — that is the test for whether it belongs on the list. They are published because a studio that writes its rules down can be held to them, and because most of them are checkable from outside.
- 01
Source art is rendered at print resolution, and kept.
Nothing is generated at delivery size and no master is deleted after export. Re-cutting a symbol two years from now has to cost an afternoon, not a re-render.
- 02
The maths is finished before the art starts.
A feature that cannot be paid for is cancelled while it is still a document — the only point at which cancelling one is cheap.
- 03
No number is published unless it is certified.
Where a value is not certified yet, this site prints an em dash instead of an estimate. There are two of them in the compliance table on the partners page right now.
- 04
Presentation never decides an outcome.
The round resolves before the animation begins. Turning off every effect in the game changes what a player sees and nothing else.
- 05
The certificate is a design input.
Compliance constraints enter at the model stage, next to return and volatility, rather than arriving as revision notes after a failed submission.
- 06
Restricted-word routing is built in, not bolted on.
Every player-facing string goes through the locale layer from the first commit, including the ones nobody expects to become a problem in a new market.
- 07
Each image is encoded on its own terms.
Quality is searched per file rather than set globally, and nothing is resampled twice. Two images of identical dimensions routinely ship at different settings.
- 08
Symbols are approved at the size they are played at.
Sign-off happens at 96 pixels on a phone, not at full width on a studio display. Anything that only works large is not finished.
- 09
If the roadmap says unfinished, it is unfinished.
Progress is published per component and marked done only when it is done. Marking a component green early buys one good week and costs the next twelve.
The silhouette does the work.
Every symbol is approved twice: once at the size it was rendered, and once at the size it is played. The second approval is the one that counts. A tumbling grid gives a player a fraction of a second per symbol, so identity has to live in the outline — the detail is what rewards someone who stops and looks, not what makes the thing readable.
It is also why the masters are kept. When the same symbol is needed for a lobby tile, a print sheet or a feature card, it gets re-cut from the original render rather than upscaled from the version that shipped.
Shipping fewer titles is the strategy, not the limitation.
A modern lobby holds more games than a player could open in a year, and shelf space goes to whoever fills it. Studios releasing a title a fortnight are not behaving irrationally. On their own terms the volume strategy works, and it is worth saying so plainly before arguing with it.
What volume costs is specificity. At that pace the maths is adapted from a library rather than written, the art is bought from the same few marketplaces everyone else buys from, and certification is whatever passes on the first attempt. No individual decision there is wrong. The result is still a lobby in which hundreds of titles are mechanically distinct and visually interchangeable.
The other way round is slower and considerably less comfortable. Each title carries the full cost of being made properly: a model written from scratch and simulated rather than reused, sets rendered at a resolution no player will ever download, an engine that has to still be maintainable when the title after next ships on it. It is a poor way to fill a lobby. It is a good way to be worth a second look.
A short catalogue also removes the hiding place. With one title submitted there is no volume to average a weak release against and no adjacent hit to carry it. Every game has to hold on its own, in public, on the same lobby page as everyone else. That is a harder position to occupy and a much easier one to explain.
The work is easier to check than to describe.
The Vault holds 25 pieces at full fidelity, each placarded with what it is, where it sits in the game and what it cost to make. The partners page covers integration, certification and commercial terms in the same register as this one.