LGTM

I overall don't love the way we have these, as package variables, but that's unrelated to this PR. We could clean up the tests if we had a config struct, and had the flags all using the *Var versions pointing to that struct. The struct is the plumbed through instead of direct access to the flags. (I guess we already have such a struct for JobOptions too)

Then we aren't accessing the flags everywhere, and testing different configurations becomes setting fields in the struct, which doesn't have ordering risk, presently mitigated by re-setting all the flags.

But that's for a different PR.