The backrub step slightly perturbs the backbone; the design step seeks the lowest-energy set of rotamers given the backbone conformation. If you run design on a thousand slightly-perturbed backbones, you’ll get a diverse set of sequences from which to choose: the perturbation to the backbone alters the sequence that maximally stabilizes that backbone. If you run design a thousand times on the same backbone, then perturb the backbone, you’ll get the same sequence a thousand times (or close to it — the design algorithms don’t always converge, but failure to converge is not a reliable way of creating diversity), and then you’ll get that sequence in slightly different backbone conformations after the backbone perturbation. A pool of diverse designs is useful. A pool of the same design (unnecessarily produced a thousand times) in slightly different conformations tends not to be.
Now, you might do some backbone perturbation after design as a precursor to something else (e.g. another round of design, filtering based on whether the designed sequence uniquely favours the designed conformation, etc.), but in general, no, it doesn’t make sense, if your only steps are a backbone perturbation and a sequence design step, to do the steps in reverse order.
.
