The anwser is described at: Origin Metadata for Records | Pinnacle 21
You should describe this at the value-level, i.e. “split” your data according to the origin in different categories.
In some cases, I would even recommend to generate different datasets (for the same domain), if the data structure is pretty dependent on the origin of the data. That would make it easier for the reviewer as well as for yourself during the mapping.
It also looks as you are trying to generate your define.xml “post-SDTM” which is never a good idea. The better way is to use a mapping tool that also generates your define.xml and keeps it “in sync” with the mapping itself.