1 · Should this even be an ontology? — the 3-question rubric

The most expensive mistake in semantic web projects: building OWL when you needed SKOS. The rubric that prevents it.

0/1 done

Theory — the rubric

The setup

OpenLearn (fictional) is a public library system publishing a catalogue of free educational resources. The first proposal from a junior team was an 'OpenLearn Educational Resources Ontology' with classes :Resource, :Book, :Video, :Topic, :Author

The senior ontology engineer killed the proposal in ten minutes. Here's how.

The 3-question rubric

Ask these in order. Stop at the first NO.

Q1. Do I need to infer anything that isn't already stated?

Examples of inference:

  • 'A reportable case is a CausalityAssessment linked to a SeriousAdverseEvent…' → yes, infer.
  • 'A HighTempObservation is an observation of gearboxTemperature…' → yes, infer.
  • 'A reading-list item belongs to topic Science if its subject heading is Physics' → no — that's just SKOS skos:broader traversal, not OWL inference.

For OpenLearn: the only 'inference' anyone wants is transitive broader traversal ('show me everything under Science, including Physics and Quantum mechanics'). SKOS gives that for free via skos:broaderTransitive. No OWL needed.

Q2. Do I need to constrain incoming data shapes?

If yes but the answer to Q1 was no, you need SHACL on plain RDFstill no ontology. The shape file IS the schema.

OpenLearn: yes, we'll add :Resource sh:property [ sh:path dc:title ; sh:minCount 1 ] and similar. SHACL only, no OWL.

Q3. Is the artefact mostly terms with hierarchy, synonyms and translations, used for indexing / search / tagging?

If yes: this is a SKOS Concept Scheme, not an ontology. Putting it into OWL classes would actively break SKOS's open-world thesaurus semantics — and lose the multilingual skos:prefLabel / altLabel patterns librarians have refined for decades.

The verdict for OpenLearn

  • Q1: NO. (Only broader traversal needed.)
  • Q2: YES, but SHACL on bare RDF is enough.
  • Q3: YES. The whole project is a controlled vocabulary of subject headings used to tag resources.

SKOS only. No ontology. No OWL classes. Anything more would be ceremony tax.

Analogy

Reaching for OWL when you need SKOS is like reaching for a relational database when you need a YAML config file. You can do it. You will spend a quarter on schema migrations to capture three settings that should have lived in 12 lines of YAML.

Reflect

Most ontology projects that die in production die because they should have been SKOS. The senior engineer's job is often to kill an ontology proposal before it burns a year of someone's career.

  • Which projects in your own backlog would the 3-question rubric send to SKOS?
  • What signals make you escalate from SKOS back up to OWL?

Reading in progress · 0 of 1 activity done