However, maybe there are some drawbacks to the current methodology.
Yes, I certainly agree. This taxonomy was primarily meant to get a high-level view of what the operators in PyTorch consist of - I certainly wouldn’t even guarantee that these tags are necessarily correct 
There is some current work (cc: @anjali411 ) on providing a more structured tagging system for operators, as well as a lot of ongoing work on providing decompositions of our operator set (see Tracing with Primitives: Update 0).