Production pattern: which order is correct?

Run the OWL reasoner to materialise inferences, then validate the result with SHACL — so SHACL sees the fully inferred graph, not the bare assertions

Validate with SHACL, then run the OWL reasoner

Run the OWL reasoner to materialise inferences, then validate the result with SHACL — so SHACL sees the fully inferred graph, not the bare assertions

SHACL vs OWL — Validation vs Inference

OWL infers under OWA; SHACL validates under CWA. They are complements, not rivals.

0/3 done

Inference vs validation

Two tools, two questions

OWL answers 'what else is true?' under the Open-World Assumption. It is designed to infer new facts.
SHACL answers 'does this graph conform to my expectations?' under the Closed-World Assumption. It is designed to validate against a shape.

Concrete example

Assertion: :Person owl:Class ; rdfs:subClassOf [ owl:onProperty :hasEmail ; owl:minCardinality 1 ].

If your dataset includes :alice a :Person without :hasEmail, OWL says nothing wrong — it merely infers Alice must have some email, we just haven't been told.
SHACL with sh:minCount 1 on :hasEmail will fail validation — that's exactly the data-quality check you wanted.

The production pattern

Use OWL to define meaning and derive inferences.
Use SHACL at the ingest boundary to validate that incoming data fits the shapes your downstream services expect.
Run the reasoner first (so inferences are visible), then the validator.

Detective vs customs officer

OWL and SHACL are a detective and a customs officer. They share a vocabulary but ask opposite questions.

A detective assumes the world is bigger than the file in front of them. If the file says 'suspect was at the train station at 9 a.m.' and 'no one at the train station took the southbound train', the detective deduces that the suspect took some other train — they don't shout 'invalid evidence!' because the file doesn't mention which train. That's OWL: open-world, deriving new facts from what's known, silent about what's missing.

A customs officer assumes the form in front of them is the whole story. If box 7 ('declared value') is empty, they don't deduce 'must be something' — they stamp REJECTED and hand the form back. That's SHACL: closed-world, validating that what's present matches the expected shape, loud about what's missing.

You want both at different boundaries — the customs officer at the front gate (ingest validation), the detective in the back office (knowledge derivation). Picking one to do the other's job is the classic anti-pattern.

Tools & resources

W3C SHACL spec: https://www.w3.org/TR/shacl/
pySHACL — reference Python validator: https://github.com/RDFLib/pySHACL
TopBraid SHACL Playground — paste graph + shape, see violations live: https://shacl.org/playground/
SHACL by Example (TopBraid handbook): https://www.topquadrant.com/technology/shacl/tutorial/
« Validating RDF Data » (Labra Gayo et al., free book on ShEx + SHACL): https://book.validatingrdf.com/

Reading in progress · 0 of 3 activities done