RM outputs the data elements that are stored in the Master Source in the TITL tag of GEDCOM and the data elements that are stored in the Source Details in the PAGE tag of GEDCOM. But I'm not sure exactly how RM makes the determination of how to construct the TITL text and the PAGE text. I think it consults the source sentence templates, but source sentence templates do not necessarily place Master Source data elements to the left of Source Details data elements in the footnote sentence. Therefore, source data can become a bit mangled in the TITL and PAGE fields as compared to how it appears in RM's actual footnote sentence.
RM's source templates have sentence templates for footnote sentences, short footnote sentences, and bibliography sentences. I have wondered if they also need a TITL template and a PAGE template for GEDCOM export.
I haven't looked in detail about the additional GEDCOM that is exported when the Extra Details (RM specific) option is chosen for GEDCOM export, but it clearly provides a receiving software with much additional information about the source. But most software processes only the TITL and PAGE tags.
With an extremely split source, all the data will go into the TITL field. This is true with extremely split source templates, but is equally true if you extremely split the citation while using the free from source template. You can't tell the difference between the templates looking only at the TITL and PAGE tags.
I don't know for sure but have always assumed that an analogous procedure takes place when RM sends sourcing information through an API such as FS/FT or ancestry.com.