Everything published on this site rests on the same procedure, run identically in every game. It is written out below in enough detail that a reader can repeat any figure we print, and disagree with our conclusions while still using our numbers.
Nothing is scored before twenty-five hours of play. Every measurement is taken on a fresh save, at default settings first, and then again after the options have been configured as well as the game permits.An action is one discrete input that advances a task: a click, a key press, a controller button, a scroll of more than one screen. Hovering is not an action. Moving a mouse is not an action. We count the shortest correct route a player who knows the interface can take, not the route a new player stumbles through, because we are measuring the design rather than the learning curve.
Three tasks are counted in every game: comparing two comparable items, performing the single most frequent operation the game asks for, and reaching the setting a player is most likely to want to change. Each is measured across forty trials and the median is published. Where a game loses a scroll position or a filter state between steps, the recovery is counted, because the player pays for it.
Every HUD and menu is examined at 1920×1080 and 2560×1440, on the same display, from a fixed viewing distance of roughly seventy centimetres. We record the smallest text the game expects you to read routinely, whether it scales with resolution, and whether an option exists to enlarge it.
A game fails on legibility when it requires reading text below our threshold in order to make a decision it asks for repeatedly. A single small label on a codex page is not a failure; a small label on the comparison screen is.
This is the measure we weight most heavily against a game. If a system operates on a quantity, that quantity should be visible somewhere. Concealing a plot point is storytelling; concealing a movement penalty that the engine is applying continuously is asking the player to guess at arithmetic the game is already doing.
We establish hidden values by experiment where we can, publish the method we used, and state plainly where we could not determine something. A game is not penalised for being complicated. It is penalised for running on figures it will not print.
Every setting a game offers is recorded: text size steps, colour palettes, HUD scaling, subtitle controls, hold-to-press substitutions, camera shake, and full input rebinding. Rebinding is recorded separately from everything else, because a game that cannot be remapped cannot be operated at all by a substantial number of the people who bought it.
Menu depth is counted as the number of screens between the main menu and the setting in question. Seven is the deepest we have recorded. Depth is a reasonable proxy for whether the settings were used by anybody outside the team that shipped them.
One number, weighted as follows: legibility thirty per cent, action cost twenty-five, information honesty twenty-five, options depth twenty. The weighting is published on every review page so nobody has to reverse-engineer it, and it does not change between games.
A score here describes the interface and nothing else. A game can be excellent and score badly, which happens regularly, and where that is the case the review says so in the first paragraph rather than burying it.
Performance, frame rate, art direction, writing, level design, sound and value for money. All of those matter and none of them is what this publication is for. We also do not cover mobile or console interfaces, because our measurements are taken on a desktop display at a fixed distance and would not transfer honestly.
Every measurement is taken on an unmodified installation. Interface mods frequently fix the exact problems we are documenting, and a community fix does not excuse a shipped design. Where a well-known mod exists we may mention its existence, but no score is adjusted for it.
One tester, one display, one pair of hands, one seating position. Action counts are objective and reproducible; legibility thresholds depend on eyesight and on the distance we have standardised, and somebody sitting closer will disagree with us. Twenty-five hours is enough to learn a menu but not always enough to find every screen in a very large game.
We also choose which games to measure, and that choice is unsystematic. There is a bias toward games with dense interfaces, because they are more interesting to measure, and that bias should be kept in mind when reading any figure described as a median.
Every game measured here was bought at retail. No developer, publisher or platform has supplied a copy, funded a piece, or seen anything before publication. There is no advertising, no sponsorship, no affiliate arrangement and no storefront link anywhere on this site.
Action counts and legibility figures are corrected in place, with a note recording what changed, whenever a reader demonstrates an error or a patch alters the interface. Where a patch improves something we have criticised, the review is annotated rather than quietly rewritten, because the original state was real and the record should show it.
There is none. This site carries no address, no form and no submission route of any kind, deliberately rather than by oversight. Nothing can be pitched here, and since every figure requires twenty-five hours of play by the person writing it, nothing could be.
