Dare Obasanjo quotes Tim Ewald’s comment
Glad to see this, because it gets us to the start of where Schematron is useful: if the XSD schema just expresses only the most invariant of the invariants of the schema, then what do we use to express the constraints appropriate in t particular phases of a document’t life or history or evolution? The “contract” is just one kind of phase.
There was some debate (Roger Costello was a champion) of default openness in XSD. But I have seen increasing use of the idea that the master/base schema has to be as loose as possible: indeed, derivation by restriction is fragile and impractical without it, because if you need a derived schema to have something optional you need to go back to the master to make it optional. A friend reports that he has taken to calling super-loose base schemas “vocabularies” and “ontologies” to avoid confusing people who expect a schema to be really prescriptive.
If the schema designer gets the user cases and they disagree in the details or only describe a set of loosely defined terms that share some semantics, there is little choice but to make the schema accept the variation. One asks how to quantify the looseness of a concept lattice. At what point does a prescription only say "object" or "thing", and the answer to that question is why XML is 'just a syntax', or minimal contract.
I hit this problem hard when working out the HumanML schema drafts. The disparate ambitions of the project heads and their wandering requirements meant that the best one could do was assemble a loosely related means of indexing the various theories of human communications. A newbie to the work called it 'monstrous' but the schema was an accurate framing of the problem.
Unless one has the power to enforce an agreement, all one can schematize is the willingness to agree.