The RosettaAntibody server has an updated antibody database compared to the release version of RosettaAntibody. Thus it is probably that the web server selected a template that was not available in the original RosettaAntibody release. From a practical standpoint, the last three residues are not very relevant for antibody-antigen interactions (in most cases). Furthermore in cases that the three residues are present the quality is not good, and in most scenarios the crystallographer simply put them in there, so the positions might not be accurate. In such cases the coordinates could be misleading. Thus if I were you, I would not read too much into those terminal residues.
Trust this helps,