I'm not sure why this discussion has popped up again, but it seems to me that there will always be people who only can grasp the bits and bytes that actually go across the wire, and there will be several sets of people who can only grasp the higher level abstractions that they see through the lens of their favorite schema aware tool.
For quite a while, the compromise seems to have been that the format itself will be defined in terms of what goes over the wire, and that any number of non-normative mappings could be included in the specification. (Non-normative only in the sense that should there ever prove to be an inconsistency, the language in the format specification should prevail).
In the case of RDF, there exists a standard means to associate a document with a mapping. This standard is called GRDDL. [1]
It defines two mechanisms for associating a document with a stylesheet. A special case (useful for XHTML) is to define a link with a rel of "transformation" to associate this document with an XSL stylesheet. It would seem to me that such an approach would work well with Atom too.
A more general mechanism is to define a grddl:interpreter attribute on the root element. Such an approach would work well too.
The way I see this working in practice is that should anybody ever actually develop a tool that is capable of doing "partial understanding" in a meaningful and useful manner, there would be a rush on the part of tool developers to add this one element or attribute to any feeds that employ an extension.
Meanwhile, it would not be harmful to mention this one element or attribute (anybody have a preference) in the specification.
- Sam Ruby
[1] http://www.w3.org/TR/grddl/
