I figured as long as the calculated gradient and energy were consistent with each other, everything would be fine. The claim of it being halved on the documentation page is what threw me off originally when I went to verify the numbers generated by NWChem by hand.