I think we've somewhat known about this, but it can crop up in unwanted ways. Currently, even with the double relaxation, we don't update the kpts for the new volume in between relaxation runs. The result is that while the volume may stop changing from one relaxation to the next, the volume change during the relaxation makes it so the kpts is not appropriate, leading to bad energies, stresses, forces, etc.
We're partially dealing with this in Emmet by having task validation, but the Structure Opt Firework should probably do some extra legwork to make sure everything is converged OR some extra validation.
I think we've somewhat known about this, but it can crop up in unwanted ways. Currently, even with the double relaxation, we don't update the kpts for the new volume in between relaxation runs. The result is that while the volume may stop changing from one relaxation to the next, the volume change during the relaxation makes it so the kpts is not appropriate, leading to bad energies, stresses, forces, etc.
We're partially dealing with this in Emmet by having task validation, but the Structure Opt Firework should probably do some extra legwork to make sure everything is converged OR some extra validation.