Scroll test that catches the notch that jumps back

Because they answer in different units, and the event says which one it used. Every wheel event carries a deltaMode field: 0 means the numbers are CSS pixels, 1 means lines of text, and 2 means whole pages. Chrome and Safari nearly always pick pixels; Firefox on Windows and Linux picks lines and reports three of them per notch, matching the Windows setting that decides how far one click of a wheel scrolls. The table on this page prints both — what your browser sent, and the same movement converted to pixels — so a reading from one browser can be compared with a reading from another.

No, that is exactly what a trackpad looks like from inside a browser. It has no detents and nothing to click against, so it reports a continuous stream of small pixel deltas — often fractional, often sixty or more a second — and then keeps sending them for a second or two after your fingers leave the glass while the momentum decays. The page detects that there is no repeating step size, says so, and switches its fault checks off rather than reporting hundreds of direction changes that are just you rubber-banding at the top of a list.

It means the encoder produced nothing for one or more of the detents your fingers felt, which is the most common way a wheel dies. The wheel body still clicks — that is a sprung arm riding over a toothed ring, and it is purely mechanical — while the part that actually reports movement has stopped keeping up. Repeat the counted run three or four times before you conclude anything, because a single miscount is more likely to be your own counting, and check whether the losses cluster at one point in the wheel's rotation.

That is a setting rather than a fault, and it lives outside the browser. Windows has a wheel-scroll figure that defaults to three lines and is adjustable up to a whole screen; macOS and most Linux desktops have their own equivalent. A setting applies uniformly to every notch you turn, so it shows up here as a perfectly consistent step size and no fault rows at all — whereas the doubled step this page flags appears in a run where every other step was half the size, which is the difference that matters.

Only if it goes the wrong way every single time. Direction inversion is a checkbox — natural or reverse scrolling — and a checkbox cannot be intermittent: with it on, up becomes down for every notch, on every application, permanently. What this page hunts is the opposite pattern, a single step that contradicts the fifty around it, which no setting can produce and which points straight at the encoder misreading one of its own transitions.

Because the pad cancels the scroll it would otherwise cause, and it has to be registered in a particular way to be allowed to. A listener declared as passive is not permitted to cancel anything, so the wheel handler here is registered non-passive on purpose — otherwise the count would slide up the screen as you built it and you would spend the test chasing the pad. The rest of the page scrolls normally; move the pointer off the pad and the wheel behaves as it always does.

Partly, and the page will tell you which part. In free-spin mode the detent arm is lifted away, so there is nothing to count and the step sizes stop repeating; the same is true of the high-resolution mode that several mice now use even when notched, where one physical click is delivered as several small events instead of one. Direction, units, totals and the raw event stream all remain meaningful. Detent counting and the two fault checks do not, and are switched off rather than left to produce numbers nobody could act on.