I love the fact that CSS is cascading; inheritance is a great idea, "where it fits". The main reason I like the cascading in CSS is because it is inherently a tree, and when it comes to parsing documents (using context free or non context free languages) a tree is very often a good fit. That and humans have long used trees (abstract or otherwise) to organise knowledge.
And since documents "fit" trees way more often than not; I'd say CSS is a good fit.
The issue is you never really know which property "cascades" or not. There are special cases for block elements, flex elements, flex children elements, sometimes even the boring "a" link object. It's basically trial and error until you somehow remember the most common cases.