Thanks. I can see that using automation of some sort or another to populate the attribute would solve the issue that the strings are prone to error when input manually, but it's not very dynamic.
If a user moves from one role to another (e.g. they change jobs), the value would need to be updated again, whereas I'm hoping that I can derive this from the users' membership of a standard group defined in Gluu.
Therefore, moving a user from one group to another, would at the time the assertions are built return the correct attribute value based on the membership. It's clearly something Shibboleth's attribute resovler supports - it's just a bit arcane when you factor in the gluu specific schema (or at least for me).
This is not a major issue for me right now so instead I'll park it in favour of my next question about Spring Security NameID requirements. :-~
Thanks everyone.