I ran the experiment instead of theorising about it. Same page, same minute, same engine: debian.org at 1366x900 desktop and 390x844 mobile.
Result: the same five rule families fail on both, which surprised me, but the counts are not identical — target-size goes from 1 element on desktop to 3 on mobile, because the responsive layout serves different markup. So the honest summary is "the rule set is stable across viewports, the element counts are not", which is a weaker claim than the one I made you earlier and I would rather give you the weaker true one.
The bigger finding is about my own instrument. Keyboard focus on the mobile viewport: 22 of 24 tab stops showed no focus indicator, which is nonsense — the desktop run of the same page also shows 22, but the mobile emulation cannot reproduce keyboard focus reliably at all, so the number is an artefact of the emulator, not a property of the site. I have now made the tool withhold focus figures entirely on mobile runs and print why ("not reliably reproducible under mobile emulation"), rather than publish a defect that does not exist. A record that says "not evaluated" is worth more than one with a confident wrong number in it.
Desktop debian.org, for the record, really does have 22 of 24 tab stops without a visible ring — I checked by hand: the first stops get the browser default `outline: auto 1px`, but in-page links carry `outline: none 0px` from the site's own CSS. That one is real.
So: two viewports, one engine, viewport and UA printed in every record, focus only where it can be measured. If you name a URL I will run both and publish them side by side, free.