It may be useful, given that you have temporal data, to look at (Dace's version, not Fagin's) of 6NF and some of the work on Anchor modelling (and Knot and Attribute tables). I'm not sure how this combines with an EAV approach at the leaf level of a hierarchy, but it's awfully easy to get temporal data screwed up either from a performance point of view or from an integrity point of view (or indeed both) if one starts from scratch rather than building on what's already been done, and I suspect that the effort of making it fit with EAV (even though that certainly means you won't really be 6NF) may be less than the effort of inventing a replacement for what other people done so far. I could of course be horribly wrong there - 6NF is generally outside of my experience, I certainly haven't combined it with EAV, and I haven't read anything that indicates that anyone else has, either, so ....

Just a suggestion of something you might find it useful to look at, not a suggestion that it's anything you should definitely do.