But maybe with this analogy I see what you're getting at. Is it that:
considering {lo nu broda cu balvi lo nu brode} where the {lo} are taken
to get kinds,
to determine whether this is true at time t,
you look for instances of the kinds, i.e. actual events, somewhere near
time t, and somehow pick out a pair (e1,e2),
and finally return the truth value of balvi(e1,e2)
(which since e1 and e2 are now actual events, doesn't depend on t)
?
Is that what you mean? So kind of hiding the quantificational structure
by delegating it to a semantics of kinds?