None of that contradicts my point that CSS generates a hierarchy of rectangular boxes from SGML/HTML/XML elements. And every system has some form of margin control, and most have some degree of support for negative margins and margin-collapse. I appreciate it's yet another thing to learn, but it doesn't violate the simplicity of the "everything is a rectangle" model that CSS uses: consider other systems that special-case their layout mechanisms (like how WPF/XAML has `<StackPanel>` for layout and `<Border>` for margin control, rather than allowing any element to have its own margins and any element to control the layout of its immediate children): it's hard to say WPF/XAML's system is "better" because rather than having a simple initial model with special-cases laid-on-top, instead you have a model comprised of only special-cases.
> CSS absolutely does disagree with our brains because it's trying to take a tree of elements with their own properties that interact based on their tree relationship and projecting that into 2D.
There isn't really any feasible alternative[1] that doesn't introduce more complexity. All the other tree-based (or hierarchy-based) systems for representing a visual document that I have experience with, such as WPF/XAML/UWP, WinForms, Qt, and others, all rely on some degree of "inheritance" from parent nodes to child nodes (and ancestor to descendant nodes) in a document (e.g. in WinForms a Control's background color defaults to its parent container's background color. Controls inherit their parents' Font, but only during initilization, thus introducing another "gotcha"), I agree that CSS shares a long list of exceptions and special-cases, but CSS's special-cases are built-on a better general-case.
If there's a better way of doing things - please share!
[1] The alternative would be something that freely translates hierarchical documents into some kind of unstructured or semi-structured "rendered document" in a separate render-space - but I don't see how that frees people from having to deal with complexity, on the contrary: it introduces even more complexity (because now the "stylesheet" is concerned with mapping from one space to another, possibly in a bi-directional manner too), whereas CSS doesn't require any significant space-translation[2]
[2] CSS's `grid` and ability to re-order `flex` elemenets, and the `absolute` and `fixed` positioning schemes do introduce new space concerns, but besides that, 95% of structured document content retains its spatial relationships when mapped by CSS's visual formatting model.
CSS absolutely does disagree with our brains because it's trying to take a tree of elements with their own properties that interact based on their tree relationship and projecting that into 2D.