In my example, the key criteria was "is there an associated common pathway model?"

Whether the meds work or not cannot tell us about whether a trait is real, but it can lend evidence to the idea if we see that there is a treatment effect and it is not heterogeneous (ex: jamanetwork.com/journal…). Homogeneous treatment effects are unrealistic for many traits because treatment is generally not targeted at the etiology of a disease comprehensively, so this can only be used to provide positive, not negative evidence wrt whether a psychiatric diagnosis tags an actual trait.

The concerns raised by PercyPrior have to do with measurement invariance, which is often tenuous, rather than etiology, which I was concerned with.