The duplication is eventually detected later by the proto schema parser, but at this moment the broken schema is already registered and its status is ERROR. Would be better to detect this early during the analysis of the annotations.