Modeled Is Not the Same as Usable

An ontology can technically model all stated requirements while still being unusable in practice.

A competency question is “modeled” if the ontology contains the classes and properties needed to write a SPARQL query that verifies it. But an ontology with 40% superfluous elements, circular references, inconsistent naming, and malformed restrictions technically passes the modeling test while failing the usability test.

The distinction matters for evaluation: automated metrics that check whether requirements are covered will miss structural problems that make the ontology painful to use. An ontology with everything needed plus a bunch of noise scores the same as a clean, minimal design.

This is why multi-dimensional evaluation matters. Coverage metrics answer “does it have what we need?” Structural metrics answer “does it only have what we need?” Expert evaluation answers “would I actually want to use this?”

Related: [None yet]