Hi Folks,
Thank you for the thoughtful responses--there's been a very interesting
convergence of ideas that I hadn't anticipated when I sent the paper.
What struck me is that three different perspectives seem to be describing
variations of the same underlying architectural pattern:

  *   XML/XSLT/XProc (the focus of my paper):
declarative pipelines, transformations, and queries over tree-structured data
  *   RDF/SPARQL (as described by Martynas):
projection + transformation over graph-structured data, with URIs serving as
first-class identifiers
  *   Markdown/Databooks (as described by Kurt):
documents combining data, metadata, and executable content, increasingly
driven by AI-based workflows
Despite the differences in technologies and data models, they appear to share
a common structure:
Representation = Transformation(Projection(Data))
or, more generally:
Output = Transformation(Data subset)
What also seems to be common across these approaches is a shift in where
application logic lives:

  *   away from imperative orchestration in code
  *   toward declarative specifications in data, documents, or pipelines
In that sense, the question may not be "XML vs RDF vs something else," but
rather:
What is the most effective way to make data, structure, and transformation the
primary organizing principles of an application?
I'm starting to think that these are not competing approaches, but different
manifestations of a broader move toward declarative, data-driven systems.
I may try to write this up more formally, but I wanted to share the
observation in case others see the same pattern (or disagree with it).
Best,
Roger