One studio can deliver the art, the maths and the client for a commissioned slot, and the publisher can still be the only party holding a licence obligation that the work has to satisfy. That is not a procurement detail. It is the reason QA on a commissioned game does not sit in one place, and why a studio that answers “we test everything internally” is answering a different question from the one being asked.
In Great Britain the split is published. The Gambling Commission’s testing strategy sets out which parts of a game a licensee must have assessed by an approved test house and which parts the licensee may test itself. The line is drawn by risk, not by discipline: the fairness of the outcome always leaves the building, while the surface the player touches — display, information, responsible-gambling controls, service-interruption behaviour — stays with the licensee, provided it can show the process behind the testing.
Read that way, a commissioned game has four columns of evidence, and the studio’s org chart has nothing to do with where they sit.

The licence does not move when you outsource
Licence condition 2.3.1 requires gambling software and remote operating licence holders to meet the remote technical standards (RTS) and the Commission’s requirements on the timing and procedures for testing. The RTS covers two areas: technical standards for how remote gambling is offered — the fairness of games, player account functionality, information provision — and security requirements drawn from ISO/IEC 27001:2013.
Two consequences follow for a publisher buying a commissioned game.
The first is that the obligation is attached to the licence, not to the invoice. When a licensee relies on a business-to-business supplier for games, the practical route is a games register reference issued by that supplier, which links the test reports into the licensee’s game register. The supplier can carry the work and even hold the relationship with the test house, but the register entry, and the assurance it represents, is the licensee’s.
The second is that “we tested it” is a claim about a process, not a result. The testing strategy allows a licensee to test certain requirements itself only if it adheres to stated good practice in development, testing and release; where it does not conform, those tests must be carried out by an approved test house instead. That single sentence is the hinge of the whole ownership question. Internal testing is not a right. It is a conditional permission, and the condition is documentary.
Column one: what the test house always takes
The part of a slot that never stays in-house is the fairness of the outcome.
For RNG-driven products, an approved test house performs statistical analysis of the RNG and its outputs before release, including any scaling and mapping inside the RNG. For the game itself, the test house examines it through maths verification, source code analysis and gameplay, to establish that it operates in accordance with the published rules. The published scope of game testing is worth reading closely, because it is broader than a maths check: verification of game design including maths, artwork and rules as displayed to players, and theoretical RTP; then software testing in an environment reflecting the intended live environment, verifying the rules and the actual RTP through simulation, emulation and manual testing, including the scaling and mapping that converts raw RNG output into outcomes.
Two things in that sentence are routinely missed when a commission is scoped.
Artwork and rules as displayed to the player are inside the fairness assessment. The display is not decoration sitting beside the maths; it is one of the three things the test house verifies. A symbol that is drawn in a way that misrepresents a paytable, or a rules panel that describes a feature differently from how the software processes it, is a fairness finding rather than a design preference. The same applies to a play-for-free version: a free game that does not implement the same rules as its play-for-money counterpart is a specific requirement of its own.
Live operation is inside it too. Progressive jackpot designs and trigger functionality are verified against the stated RTP, and licensees must monitor actual against advertised RTP for both under- and overpayment, with an approved test house assessing those monitoring measures annually.
Column two: what the licensee tests itself, and under what conditions
Everything a player reads, configures or depends on when something goes wrong is in the licensee-testing column: the accessibility of rules and site information, stake and session controls, autoplay restrictions, the behaviour of multi-state games during a service interruption, and protection against cheating and collusion where technical solutions are used.
This is where a publisher’s in-house or outsourced QA actually sits. The studio builds the client, the session tools and the interruption behaviour, but the obligation to test them before release belongs to the licensee, and the permission to do that testing depends on process evidence that most studios have never been asked to produce.
The strategy names the conditions. Development: source code in a secure environment with an audit log of access and retired versions retained with the dates they were retired; access limited to what the job requires; technical documentation suitable for independent review; source legible and structured enough for static analysis; write access to platform source withheld from people working only on game-specific development; and changes to critical modules peer-reviewed by developers who are independent of the change.
Testing: logically separate development and test environments, testing performed by staff other than the developers, and an independent assessment that the changes made are the changes documented — the strategy suggests file comparison as the mechanism.
Read against a small studio’s reality, that is a demanding list, and it is the list a publisher should be asking about before the first build, not after certification. The change-control analysis sorts which of these surfaces a platform review will ask about, and the post-approval scope brief covers what happens to them after release.
Column three: the engineering questions that never become test-house questions
A large share of engineering QA never reaches a test house at all, and this is where a publisher loses visibility.
New channels are the clearest case. When an existing game is released on a new channel, the channel must be tested by an approved test house, but the published scope of that testing is limited to the user interface and player display — a manual test of how the game client presents results — with reference to the original game test report, provided the backend design and functionality were not altered to accommodate it. In practice, most of the engineering evidence for a new channel is the studio’s client work, and the test house confirms the display.
Platform updates are the second case. Updates to third-party client operating systems and browsers generally do not require external retesting, and the expectation is that the licensee’s own testing confirms satisfactory performance when new client operating system and browser versions are released. That distinction matters: an update to the RGS or the RNG underneath the game engine can require retesting, while an iOS or browser release does not, and the two are often conflated in release planning.
The third is the device set. Where a game is designed to work across a variety of devices or browsers, testing should cover the most commonly used ones, and those devices and browsers have to be identified in the test report. A studio that tests on a modern handset and an emulator has not necessarily tested what the report says it tested. The mobile performance budget sets out how to make that device set and its thresholds falsifiable, and the aggregator integration tests are the acceptance evidence for the interface layer a studio does not own.
Column four: the record that decides who is right later
The strategy’s change-documentation requirement is the artifact that ties the first three columns together, and it is the part of QA a publisher is least likely to have specified.
Accompanying any RNG or game change, the documentation must record a unique change ID, the game number or RNG identifier, the delivery channel or channels, a description of the change, whether it is classified as major or minor, the justification for that classification, confirmation of internal testing and documentation for minor changes, confirmation of adequate external test-house assessment for major changes, and the relevant manager’s authorisation.
The classification rule is the commercial edge. A major update is any software change that may affect the fairness of a game, and fairness includes the RNG, scaling and mapping, and the game rules including how the rules are processed by the software. The published examples are instructive in both directions. Rebuilding the symbol arrays to cut CPU load is a major update, because although no rule changed, the software implementation of the rules did. Correcting a bonus-round win calculation for a rarely encountered scenario is a major update. Against those sit a sound-format change to support an older mobile operating system, a hat colour and background graphics changed because a licence expired, a bonus round displayed incorrectly on screen, and a stake-selection dialog that does not render fully in one browser.
Those minor examples are exactly the changes an art or front-end team makes late in a project, and each one is a documented decision rather than a bug fix. The strategy is also explicit that the boundary is a judgement call: its own example note warns that a bundle of small visual fixes could easily fall into the major definition, and that where doubt exists the original test laboratory should be consulted.
What this means when you are buying the game
Ask for the row owners, not the org chart. Four questions cover the split.
Which columns are yours and which are mine? The test-house column is always the licensee’s obligation, even when the supplier holds the relationship. Say so in the proposal, and name the applicant of record.
Who signs each artifact? For the four columns that means, at minimum, the test report and its open findings, the internal test evidence for the licensee-testing requirements, the client and channel test evidence, and the change record. An artifact with no named signatory is an artifact nobody will produce under pressure.
What process evidence will you hold, and can I audit it? Source access control, retired-version retention, peer review of critical modules, separate test environments and an independent assessment of the diff are the conditions behind the in-house permission. A publisher that cannot inspect them cannot rely on the in-house column.
What happens at the boundary between columns? A defect found in gameplay that turns out to be an art misrepresentation, or a display failure that is really a rules-processing failure, is the normal case rather than the exception. The rules specification guide and the frontend acceptance criteria are the two documents that make that boundary testable, and the animation handoff is the part of it that most often arrives without evidence.
None of this requires a publisher to distrust its supplier. It requires the proposal to say which evidence each party produces, because the assurance is not transferable even when the work is. The proposal comparison method normalises those lines across suppliers, and the commissioning guide covers how the brief is assembled before any of this is quoted.
If you want the four columns, the signatories and the change record settled as part of the build scope rather than discovered during submission, request a project quote.
Questions buyers ask
Does outsourcing development transfer responsibility for testing?
No. In Great Britain the technical standards and the testing requirements attach to the licence, so the licensee remains responsible for ensuring new products are adequately tested before release and for supplying the test report evidence. A business-to-business supplier can hold the relationship with the test house and issue the games register reference that links the reports, but the obligation and the register entry stay with the licensee. A commission should therefore state which party produces which evidence rather than assuming the studio’s internal testing discharges it.
What can a licensed operator test internally instead of at a test house?
The requirements that a licensee may test itself are the player-facing ones: information and rules availability, stake and session controls, autoplay restrictions, service-interruption behaviour, play-for-free parity, and anti-collusion measures where technical solutions are used. That permission is conditional, not automatic. It depends on documented good practice in development, testing and release — controlled source access with a retained audit log, separate test environments, testing by staff other than the developers, peer review of critical modules, and an independent check that the changes made match the changes documented. Where a licensee does not conform, those tests have to go to an approved test house.
Is artwork part of the fairness testing?
Yes, as displayed to the player. The published scope of game testing covers verification of game design — maths, the artwork and rules as displayed, and the theoretical return — followed by software testing that verifies the rules and the actual return through simulation, emulation and manual testing. A display that misrepresents a paytable or a rules panel that describes a feature differently from how the software processes it is therefore a fairness question, not a cosmetic one.
Does a new platform or browser release mean the game must be retested?
Generally not, and the distinction is worth writing into the release plan. Updates to third-party client operating systems and browsers are expected to be covered by the licensee’s own testing confirming satisfactory performance, while an update to the RGS or the RNG that underpins the game engine can require retesting. A new delivery channel is different again: the channel itself must be tested by an approved test house, with the scope normally limited to the user interface and player display.
How do we keep control of changes after launch?
Make the classification rule part of the contract rather than a matter of opinion, and require the change record to be produced with every update. The record carries a unique change ID, the game or RNG identifier, the channels, a description, the major or minor classification with its justification, confirmation of internal testing or external assessment, and a manager’s authorisation. Any change that may affect fairness — the RNG, scaling and mapping, or the game rules including how the software processes them — is major, and even a performance refactor of the symbol arrays falls inside that definition.









































