Point taken Eduardo: there does seem to be wide consensus that separation of model from presentation is a Good Thing. That's why it's ironic that in the XML community we're always talking about how our document instances should be readable . Readablility is a goodness measurement of presentation right? A model wouldn't have to be readable right? There are a whole range of tradeoffs possible. A presentation-free model exists at one end... pure presentation details is at the other. Inasmuch as we've already chosen a midpoint, I see nothing absurd about discussing possible surrounding midpoints. In order from presentation-free to presentation-full here are some points in the continuum (choices we could make for our document/message meta-structure): * a binary structure -- you've got to read the schema (not included) to understand the structure * a text-encoded structure that's fairly readable but which carries no markup save positional and nesting markup (think LISP S-EXPR's) -- you've still got to read a schema to know what's going on * some sub-XML type thing that's e.g. less verbose -- now you have tags but perhaps structure end is a generic close delimiter -- not a repeated tag name. * XML -- now you've got tag names, but in what language -- you've already picked a presentation right? * XML with human readable code list values -- again, in what language If we're going to go for a clean separation of model from presentation the I'd guess that we'd want to e.g. use numbers to identify code list values in a language-neutral way. That being said I find it a little bit of a double standard, given that we've chosen XML as a basis (what with all it's redundancy-for-the-sake-of-readability) I just re-read Phil's message: > > The numeric representation of the country identifier > > is the only part that we rely on in canonical XML > > markup based on the ASN.1 schema. Only the numeric > > portion of an abstract value such as ROM(642) > > need ever appear in the transfer syntax where there > > is typically no human reader involved. I think the key word is rely . The way I would interpret this idea for UBL would be that we would make the pure identifier (the numeric one) required, and we'd make the textual one ( ROU ) optional. Seems like we ought to give some guidance that says: A UBL-compliant processor will operate only on the pure identifiers -- not on the textual ones . In what artifact could we make such a statement? Is that part of our charter? I think it should be. >