What the page is counting
Every notch or trackpad swipe that lands in the box becomes one WheelEvent. MDN describes deltaY as the amount of scroll on the vertical axis and deltaMode as the unit for that amount: 0 for pixels, 1 for lines, and 2 for pages. On most discrete mouse wheels the browser fires one event per physical notch, so a clean wheel builds a stack of similar |Δy| values. This page keeps those events in memory for the session, splits them into up (negative deltaY in the usual browser convention) and down, and derives a median |Δy| from the same list.
A tick is marked as a skip when its |Δy| is at least 2.5 times that median. A single missed notch that merely doubles the usual line-mode step (100 → 200) stays under the line; a 3x to 4x jump crosses it. Dirt under the encoder, a worn ratchet, and some OS smooth-scroll paths can all produce that signature, so the count is a clue rather than a bench verdict. Nothing is uploaded: the numbers live in this tab until you reset or close it.
How to read direction, speed, and delta mode
Roll both ways. A healthy session should show up and down counts moving together when you reverse direction. If only one direction registers after dozens of ticks, the OS may have inverted natural scrolling, or the wheel switch may be failing on one side. Speed is ticks per second across the span from the first to the last event in the session, useful for comparing a light flick to a heavy one, not as a lab accuracy claim.
Delta mode tells you which unit the browser is using for Δy. Line mode (1) is common on desktop mice; pixel mode (0) shows up more often on trackpads and smooth-scroll paths. Comparing median |Δy| across modes is not meaningful, which is why the page prints the dominant mode beside the median instead of mixing them into one notch-size number.
What this scroll test cannot prove
It cannot see the hardware report rate of the wheel sensor, and it cannot separate a dirty encoder from an OS accessibility setting that rewrites scroll deltas. It also shares the page with any other program that generates wheel events while the pointer is over the box. If you need button mapping, double-click timing, or a polling-rate estimate, use the button and double-click checks on this site; if clicks feel late because the display is in a low-Hz mode, the Hz measurement page measures that ceiling.