Most commissioned slot projects do not fail at the end. They fail at a point in the middle where everybody believed a phase was finished and nobody could say what finished meant. The maths was “basically done”. The art was “approved in the last call”. The build was “working on my machine”. Six weeks later the same three sentences are being argued about with a payment attached to each.
The structure that prevents this is unglamorous: divide the commission into stages, and let each stage close only on an artefact that survives being looked at by someone who was not in the room. That is what a stage gate is. It is not a meeting, and it is not a status report.

What a gate has to contain
A gate that works answers four questions, and the answers belong in the engagement document rather than in a slide.
- What closes it. One named artefact, in a defined format, at a defined version. “The prototype” is not an artefact; “the prototype build plus its simulation output over a stated number of spins, at a stated version” is.
- Who signs it. A named role on each side, not a committee. The signer accepts the artefact, not the project.
- What it proves. The narrow claim the artefact supports — that the mechanic pays what the brief said, that the client presents the approved outcomes, that the rules text matches the implemented behaviour.
- What it cannot prove. Every gate has a blind spot, and writing it down is what stops the same artefact being used later to defend a claim it was never evidence for.
The fourth field is the one teams skip, and it is the one that does the most work. A gate whose limits are stated in advance cannot be reinterpreted six weeks later.
Gate one: the mechanic, measured
The first gate is the cheapest place to lose a concept. A prototype exists to answer a behavioural question about the feature — what triggers it, how often, what the player decides inside it, and what the distribution of outcomes looks like — not to look finished.
It closes on a build a stranger can run plus the numbers the build produced. The prototype gate sets out what belongs in that validation, including the difference between a prototype that demonstrates a feature and one that measures it.
What the gate proves: the feature behaves as described and its frequency is known. What it cannot prove: that the feature is enjoyable, or that its frequency is commercially right. Both are judgement calls made later, with better information and worse options.
Gate two: the maths, as a document
Maths is where a commission acquires its permanent constraints, because on Engine the approved figures are the game. Approval fixes base RTP, per-mode RTP where a title carries more than one mode, the maximum win, and the standard deviation of a single round for the approved maths. Once a game is live the maths is locked; a change is accepted only to fix a confirmed bug, and a new mode or a mechanic change needs a new title with its own submission.
That makes this gate different from the others: it is the last point at which the maths is still cheap to move. It closes on the package rather than on a conversation — the par sheet, the simulation evidence behind each mode, and the mode index. The maths model requirements describes what that package contains and how simulation evidence is separated from a claim about it.
Two design constraints belong at this gate, because after it they become expensive: Engine games are stateless, each bet independent of the last with no jackpots, no gamble-continuation and no early cash-out; and a game carries at most twenty modes, with exceptions handled through support. A mode index that has to grow after approval does not grow.
What the gate proves: the numbers are reproducible and the mode set is complete. What it cannot prove: that the numbers will perform commercially, and that the client will present them correctly — which is the fourth gate, and the reason that gate has its own evidence.
Gate three: art and animation, as an inventory
Art is the phase most often closed by atmosphere. A style frame is approved, the palette feels right, and production begins on a symbol set whose size nobody has counted.
The gate should close on an inventory: every symbol at production resolution, every state a symbol can be in, every transition the feature needs, and the animation list with its frame budgets. The animation brief is the shape of that document, and the reason it is a list rather than a mood board is that the list is what engineering can build against.
What the gate proves: the visual system is complete enough to implement without inventing assets mid-build. What it cannot prove: that the art is good, or that it will read at the size it will actually be seen.
Gate four: the client, on parity
The client gate is not “the game is playable”. Playability is a demo property. The gate is that the client presents the approved outcomes — that what a player sees corresponds to what the maths produced, for every mode, including the zero-win rounds and the round-restore path.
Maths-to-client parity defines the evidence: a comparison against the reference outcomes rather than a spot check of a few spins, with the differences explained or fixed. The frontend acceptance criteria cover the surrounding behaviour the parity evidence does not touch — launch parameters, bet levels, active-round restore, insufficient balance, the rules screen.
What the gate proves: the build and the maths agree. What it cannot prove: that the build performs, on the devices the players actually hold. That is the next gate, and separating them is deliberate.
Gate five: the declared device matrix
Performance is the gate that most often has no pass condition, because nobody wrote down the devices. A build that runs smoothly on a development machine and a current flagship phone can fail on the hardware a meaningful share of players use.
This gate closes on a measurement against a matrix: the browsers and device classes the operator expects to serve, the frame budget claimed for each, and the measurement conditions. The performance budget and mobile test plan sets out how that matrix is declared and what counts as evidence.
What the gate proves: the build meets a stated budget on a stated set of devices. What it cannot prove: that it meets it on the device the reviewer happens to own, or in conditions nobody has described.
Gate six: rules and player-facing copy
Every rule a player reads is a claim about the build, and every claim has to be traceable to behaviour. The gate closes when the rules text, the paytable and the in-game messaging match the implemented mechanics line by line, with no inherited sentence describing a feature this title does not have.
The rules specification guide treats that as a build and acceptance exercise rather than a copywriting task. It matters commercially as well as editorially: a rules screen that overstates a maximum win is a defect in the product, not a typo.
What the gate proves: the description and the behaviour agree at that version. What it cannot prove: that the text is clear to a first-time player. That is user research, not acceptance.
Gate seven: the submission package
The submission gate is a completeness check, and it is the gate most likely to be pulled forward into the schedule and then skipped. Its artefact is the package itself: which front build and which maths version are being submitted, the evidence that the maths source reproduces the published figures, and the record of everything that changed since the previous version.
It is also the gate where the twenty-mode ceiling stops being a design note. Because modes cannot be added after approval, the index submitted here has to contain every mode the title may need for its commercial life.
What the gate proves: the package is internally consistent and complete. What it cannot prove: that it will be approved, or on what date. Anyone who offers a platform-controlled date as a gate condition is selling certainty nobody holds.
Gate eight: the localised build, if there is one
The final gate is not a translation gate. In a licensed market a localised build is a certified build, so the language, currency, stake ladder and messaging are part of what has to be accepted. The localised-build evidence set works through what travels with a language change and what has to be re-established; where certification is game-level, as in Brazil, a localised variant is another game to certify.
If that is true of the markets in scope, this gate has its own artefact, its own submission and its own date. Scoping it as a copy of the first build is the mistake that turns a two-month addition into a second project.
What no gate proves
No gate in this sequence proves that the game will perform. A commission that clears every gate has a reproducible maths model, a client that presents it faithfully, a build inside a declared budget, rules that match the behaviour, and a complete package. It does not have evidence about retention, first-time deposits or revenue, and no acceptance process produces that evidence, because the comparison it would require is confounded by launch timing, promotion spend and lobby position.
It also does not replace anything on the operator’s side of the boundary: an aggregator’s own acceptance process, the operator’s reconciliation reporting, and whatever the market’s regulator requires sit outside the studio’s gates and can move independently of them.
Writing the gates into the quote
The practical value of the sequence is that it makes a proposal comparable. Comparing custom game proposals is far easier when each proposal names the same eight artefacts: the differences between suppliers stop being differences of vocabulary and become differences of what is actually delivered.
The stages also give the payment schedule something to attach to. A phase payment released at a gate releases against an artefact both sides have already defined, which is why the gate register belongs in the engagement document and not in a project plan that gets rewritten in week three.
The studio engagement route describes how a full build is structured around these stages. If you are writing the acceptance structure for a commission and want the gates defined against your platform, markets and target devices, request a project quote with the stages you already have in mind.
Questions buyers ask
What is a stage gate in a commissioned slot project?
A stage gate is a point where one phase closes on a named artefact — a measured prototype, a maths package, an asset inventory, parity evidence, a device-matrix measurement, matched rules text, a submission package — and a named person on each side accepts it. It is defined before production starts and it answers four things: what closes the gate, who signs, what the artefact proves, and what it cannot prove.
Which stage should be gated most tightly, art or maths?
The maths. On Engine the approved figures are the game: approval fixes base RTP, per-mode RTP where a title has more than one mode, the maximum win and the standard deviation of a single round, and once a game is live the maths is locked to confirmed bug fixes. Before approval the maths is still cheap to change; after it, a new mode or a mechanic change needs a new title and a new submission.
Can a game be changed after the submission gate?
Only within the narrow boundary the platform allows, which is why the mode index has to be complete at submission. A game carries at most twenty modes and modes cannot be added after approval, and a live game’s maths is not retuned for design reasons. Anything larger than that boundary is a new title with its own package.
Do acceptance gates prove a game will perform commercially?
No. They prove the build matches the brief and the maths, performs inside a declared budget on a declared set of devices, and is described accurately to players. Commercial performance is a separate question with confounded evidence, so it should be framed as a hypothesis written before launch rather than as an acceptance condition.









































