The tag system on MathOverflow is designed in such way that each question should contain top-level tag; these are the tags which correspond to subject areas on arXiv and generally have format such as gm.general-mathematics or gn.general-topology. (The first two letters are the shortcut used on arXiv and they are followed by the name of the area.)
For more details see: What are top level tags, and why should I try to use them? and Why are MO tags formatted as they are?. Quote from the latter: "As a rule, each question should have at least one of these top-level tags." Also quite recently François G. Dorais mentioned such tags: "For example, the systematic use of broad area tags borrowed from the arxiv on day one brought a lot of stability to the general tagging system here."
So as far as I can tell, the rule that at least one top-level tag should be used in each question - or at least in most of the questions - is still in place. (And it seems to be a useful rule for several reasons.)
However, it seems to me that some tags which do not have arXiv format are used basically as top-level tags. For example, it seems that linear-algebra, graph-theory, set-theory are used in this way. (Maybe also real-analysis?)
TL;DR: My question is which are the tags that do not have "arXiv format" but still can be considered top-level tags. (In the sense that they cover quite broad area and that the question which have one of these tags but no arXiv tag can be considered as tagged in accordance with recommendations in the FAQ.)
Or, if the questions from the areas mentioned above should also additionally have "arXiv " tag, which tag should be used in these cases.
EDIT: I have looked a bit through the tag-wikis and saw that the tag-info for co.combinatorics (revision history) explicitly mentions graph theory. Similarly, the tag-info for lo.logic (revision history) explicitly mentions set theory. In these two cases the two "arxiv" tags are definitely close to graph/set theory and these tags are often used simultaneously. (Out of 2807 questions tagged set-theory, 1102 are also tagged lo.logic. Out of 2947 questions tagged graph-theory, 1147 are also tagged co.combinatorics.)
Linear algebra is explicitly mentioned in the tag-info for ra.rings-and-algebras (revision history). But in this case is much less clear whether this would be a suitable top-level-tag for a typical linear algebra question. (Indeed, out of 3146 questions tagged linear-algebra, only 132 are also tagged ra.rings-and-algebra.)