Individual reviews cover one game. These notes cover patterns: which failures repeat, which are cheap to fix, and which have quietly become worse over the four years this bench has been running.
Two notes published · 212 games in the archive · four devices each
Note
Sustained holds appear in eighty-four per cent of the archive and offer a toggle alternative in nineteen per cent. The gap between those two figures is the largest single accessibility failure we measure.
Note
Before mapping comes detection. Twenty-two per cent of the archive never enumerates the device, which means no amount of rebinding will help, and no storefront page mentions it.
Across the archive, actions locked out of rebinding are three times more likely to be holds, chords or double-taps than the average action. Whatever the reason, the effect is that the hardest inputs are the ones you cannot move.
Forty-one games in the archive ignore an attempt to rebind a locked action without saying anything. Testers consistently read this as a bug and retry it, sometimes for several minutes, which is a worse outcome than a plain message.
Where a toggle alternative exists, it sits in a separate accessibility screen in seventy-eight per cent of cases rather than beside the binding it modifies. Players who need it frequently never find it.
Small teams ship complete rebinding at roughly the same rate as large ones. Red Pillars was made by three people and does everything right; several far larger games in the same bench do not.
Median peak simultaneous inputs across the archive has moved from 3.4 to 4.5 over four years. Median sustained holds has moved from 1.2 to 2.5, which is the more consequential of the two changes.
Four testers is four pairs of hands. The figures here are properties of the software — input counts, locked bindings, device detection — and those transfer exactly. Comfort does not. A scheme that four people found manageable may still be unusable for you, and the archive is deliberately built so that you can check the underlying numbers rather than trust our comfort.
Where a game's behaviour changed after a patch, the review is re-run and the date of the re-run is recorded. Twelve reviews in the archive currently carry a superseded figure with the original kept visible.
