This test asserts the behavior seemingly agreed upon through https://github.com/w3c/csswg-drafts/issues/11044#issuecomment-2486935794. This is unlikely to be the final specified behavior, hence the test is marked tentative.

Basic case: treat margins as 0.

When resolving stretch, don't treat margins as 0 if there is a sibling on that side. (Hand-wavy: we don't collapse margins through siblings for the purpose of resolving stretched heights.)

Then when actually doing margin collapsing, .inner's margins DON'T collapse with .outer's because of the usual rule that margins don't collapse through new formatting contexts.

When resolving stretch, don't treat margins as 0 if there is a sibling on that side. (Hand-wavy: we don't collapse margins through siblings for the purpose of resolving stretched heights.)

Then when actually doing margin collapsing, .inner's margins DO collapse with .outer's because they collapse through the empty siblings.

Only treat the BOTTOM margin as 0. Margins do collapse, so top edges touch.

Only treat the TOP margin as 0. Margins do collapse, so top edges touch.

Account for the margins when parent establishes a BFC.