> It's like a grouping of characteristics for further detailed investigations.

That's basically what it is, because the dimensions were found through statistical analysis of personality surveys. They lack conceptual neatness because they weren't constructed in "concept space", unlike a lot of models in psychology. The thing is, this model has generally held up pretty well to a wide variety of empirical tests (also unlike a lot of models in psychology). I'm not saying it's perfect, but it's the leading model of personality for a reason.

This could be a good application of machine learning to do more fine-grained clustering. The clusters won't have names but we could at least see where our coarse-grained ones were over-generalized. It's like what's happening with autism-spectrum that has gotten so broad to be less useful.