What do people who enable reduced motion expect to be removed: all animation, large movement or only autoplay?

Open question

The media feature has exactly two values, no-preference and reduce (MDN); it does not say how much to reduce. Authors implementing it must therefore decide how far to go. One reading removes every animation and transition, so that the page behaves like a document with instant state changes. Another keeps short, low-amplitude transitions that show cause and effect (a panel opening, a toggle switching, a list item fading out) and removes only large movement: scaling, parallax, slide-ins, scroll-linked effects and autoplaying media. A third reading treats the setting as a request for calm rather than stillness and keeps everything but the effects known to trigger vestibular symptoms.

The setting may be enabled by people with vestibular disorders, for whom the second and third readings may still leave harmful effects in place, but plausibly also by people who find motion distracting, who use assistive technology that interacts badly with animation, or who simply prefer a quieter interface. These groups may not want the same thing, and the media feature does not distinguish them. Is there published research, usability testing or documented user feedback on what people who enable the setting expect; whether "reduced" or "none" better matches that expectation; whether expectations differ by the reason for enabling it; and whether a per-site motion toggle (often recommended alongside the media query) changes the answer by giving users a finer control?

What a useful answer contains

The population studied (reason for enabling the setting, platform, assistive technology in use), the interface elements tested (state transitions, loading indicators, scroll effects, page transitions, autoplaying media), how expectations were elicited (interviews, preference tasks, diary studies, complaint logs from a product with the setting honoured), the size of any effect and its uncertainty, and the date, since platform defaults and the population using the setting may change over time. Anecdotes from individual users are welcome if labelled as such; a summary of how large products currently interpret the setting would also help, provided it is clear that it describes practice rather than user preference.