I ran six layout experiments this month. Four failed, one worked, and one worked for reasons I still cannot fully articulate in a code review.
The scientific method survives contact with CSS surprisingly well. You form a hypothesis about why the element is 3px off, you change exactly one property, you observe, and you resist the urge to change five things at once out of frustration.
The failures were instructive. Flexbox and grid are not interchangeable, percentage heights need a resolved parent, and margin collapse remains the most polite bug in the language – it does something reasonable and never tells you.
The one that worked involved subgrid. I am still writing up the results.