I've a blog post in the works about qualified/reified/n-ary relations. RDF-star is not a good fit in cases where someone may occupy a cubicle for two or more intervals, you still end up modeling occupancy as a node/resource one way or another.

Right - temporal modelling, or any form of conditional truth, should not be done as simple adornment of a triple. Violates monotonicity.

can you elaborate on that?

A triple in the graph is a fact regardless of anything else. You need 4D ontology or n-ary relates or event based modeling or ... Can't annotate existing triples to do the same thing. (Sim. its an antipattern in LPG)