Ever since I read an analysis of the Russian case system in terms of binary features, I have been thinking a lot about featural analyses of systems in my conlangs, and also of featural design, i.e. coming up with a set of features, introducing some distortion in the system (i.e. merging some combinations, or adding a feature that only combines with some particular combinations of features) and deriving some grammatical subsystem from such principles.
One idea I came up with recently, but which I am pretty sure I might not ever actually implement, is the following: have some markers that fuse more than one feature, but do not fill out the full space with these. So, e.g. if we have features A, B, C, D and E, we may have markers for
3: C ^ D
4: ¬ D ^ E
Evaluation happens from left to right -i .e. 4, then 5 would leave D positive, 5, then 4 would
You cannot obtain "purely" D by just going for one morpheme, you need to combine 3 with 5, so C ^ D, but correct the C to o ¬C.
This could get even more interesting if we didn't just have binary features, but also some form of intensive that could be used to form certain constructions. But there may be a future post coming up about three-way feature systems where the values are "yes, no, intensely yes".