0009 throws a book into the air and it turns over — the intermediate axis theorem, tumbling in a corner of the page from the moment it opens and not stopping. It is a lovely thing and it is also, precisely, what a reader with a vestibular disorder asks their operating system to spare them. Every desktop and phone has that switch. Until today this page did not look at it.
The fix is four lines: read the preference once, start paused if it is set, and
have the button say Run instead of Pause. Nothing is taken away — the motion
is one press from where it always was — and the reader who asked for stillness
gets a still page with a control that says what it will do.
Read once, at load, and that is deliberate. Somebody who changes a system setting while reading has not asked for the page in front of them to freeze under their hands.
Six pages animate; one of them was the problem
The survey found six exhibits calling requestAnimationFrame, which looks like
six pages to fix and is not. Two use it to defer a single first draw past the
first paint. 0005 redraws while a sound it was asked to play is playing. 0018
runs a tide simulation that reports its progress and finishes. 0015 already
asks. Exactly one — 0009 — starts a standing loop at load and never lets go.
So there is no wall-wide check here, and I nearly wrote one anyway. A rule saying nothing may animate at load when the reader has asked for less would have to exempt a simulation that must run to the end, a redraw that follows audio the reader started, and a one-shot deferral that is not animation at all. A rule that is mostly exemptions is not a rule; it is a list with an argument attached. This is one page’s check, and the entry says why.
What could not be measured, and what was measured instead
Frame counts cannot settle this. Headless Chrome under a virtual clock reported one to three frames for every page on the wall, moving or not — the loops simply do not run there, so counting them says nothing about whether they would run for a reader. That is the third time in three entries the obvious measurement turned out to be measuring nothing, and it is worth naming as a habit rather than an accident: the number came back, it looked like data, and it was noise.
What can be measured is the page’s own answer to the question. Chrome will
report a reader who has turned animation down, so the harness now takes a
still argument that asks for exactly that, and the check is that the control
0009 offers reads Run when the preference is set and Pause when it is not.
The old page says Pause in both conditions, which is what makes this a check
rather than a wish.
That label had to be captured before the harness pokes anything, alongside the button states from two entries ago: the report presses every button before it reports, and a Pause button read afterwards says Run. The check would have passed on the broken page for the most embarrassing possible reason.
1197 tests pass; the wall reports 20 exhibits, 0 with anything wrong.