The four devices
Every game is played to completion or to forty hours, whichever comes first, on a standard wireless gamepad. It is then re-tested on keyboard and mouse, on a flight stick where the genre makes that meaningful, and on a commercially available adaptive switch interface that presents to the system as a generic controller.
All hardware is bought outright. No device in the bench was supplied by a manufacturer, and no manufacturer has been told in advance which games are being tested.
| Device | Role in the bench |
|---|---|
| Standard gamepad | Primary playthrough and scoring |
| Keyboard and mouse | Comparison scheme, chord and hold distribution |
| Flight stick | Axis handling, genre-relevant titles only |
| Adaptive switch interface | Detection and reachability testing |
Counting rules
These rules are fixed so that figures are comparable across four years of benches. Where a rule has changed, the change is noted and affected reviews are re-run rather than adjusted.
Bound actions
Counted from the binding screen, excluding menu navigation and excluding duplicate bindings for the same action. A game that binds "interact" and "interact (hold)" separately is counted as two.
Peak simultaneous inputs
The highest number of inputs that must be active at the same instant during normal play, verified in the hardest section reached during the playthrough. Optional expert techniques are excluded unless the game's own tutorial teaches them.
Sustained holds
Any input that must be maintained for more than one second to keep a state active. Charge inputs resolving in under a second are counted separately as timing inputs, not holds.
Rebindable share
The proportion of bound actions that can be reassigned from inside the game's own menu, without editing a configuration file, installing software, or using a device-level remapping utility. Actions locked out are always named in the review, because a percentage hides which ones were lost.
Device detection
Tested before mapping. A device counts as detected only if the game lists it as an input source and allows at least one action to be assigned to it. A device visible to the operating system but absent from the game's menu is recorded as not detected.
Scoring
The score out of ten is a score for the game as a game — design, execution, how it holds up over the playthrough. The input figures sit beside it and are never blended into it, for two reasons.
First, a single number cannot carry both meanings without becoming useless for both. Second, the input figures are what a reader needs in order to decide whether the score is even relevant to them. Splinter Arena scores seven point nine and is either an excellent purchase or an impossible one depending on facts about you that no score can encode.
Measured properties of software transfer exactly from our bench to your desk. Comfort does not transfer at all. Everything on this site is written to keep those two categories visibly apart.
Testers
Each bench is run with three additional testers alongside the primary playthrough, using their own hardware and their own habits. Their notes appear in reviews with editing for length only, and they are identified by device rather than by name.
Three additional testers is not a sample. It is a check against the primary tester's assumptions, and it is described that way rather than dressed up as research.
Patches and re-runs
Where a patch changes the input layer, the bench is re-run and the review is updated with the date of the re-run. The superseded figures stay visible rather than being quietly replaced, because a game that fixed something in month eight is a different fact from a game that shipped correct.
Twelve reviews currently carry both an original and a revised set of figures. Scores are only changed if the game changed; a re-run that produces better input figures does not retroactively improve the score for anyone who bought it at launch.
What this method cannot tell you
It cannot tell you whether a scheme will be comfortable for your hands. It cannot tell you how a game behaves on hardware we do not own, and the archive says so where relevant rather than extrapolating. It cannot account for the enormous variation in what "sustained" means physically from one person to another.
What it can tell you is exactly how many inputs a game demands, which of them refuse to move, which hardware it never sees, and whether the thing standing between you and the game is the design or the settings screen. In one hundred and thirty-eight cases in this archive, it is the settings screen.
Independence
Every game is bought at retail at full price. No review copies, no early access arrangements, no advertising, no affiliate links, and no advance sight of reviews for anyone involved in making the games. Where a developer has responded to a review, the response is summarised in the review itself and the original text is left unaltered.