Case Study: Open Data Infrastructure
Publishing machine-validated open data structures for the public sector
Structured, tagged data is only valuable if its quality is machine-verifiable. Across national skills data, heritage archives and an open-source toolkit, we build formally structured data, validate every statement with deterministic machine checks, and publish both the data and the validation tooling openly so anyone can re-run the checks.
Skills England ontology
51,355
triples, 0 SHACL violations
Heritage archive
292
digitised frames, 0 SHACL violations
Validation toolkit
Open
published on GitHub, external contributors
The Challenge
Public bodies publish ever more structured and tagged data: classifications, registers, taxonomies, machine-readable records. The tagging is the easy part to claim and the hard part to trust. A dataset can look structured while carrying broken references, malformed records and statements that violate its own schema, and none of that is visible to a human reader. Quality assertions ("the data is clean") are not the same as quality evidence.
The fix is to make quality machine-verifiable: publish the data with a formal schema and constraint rules, validate every statement against them, and publish the validation tooling itself so the checks can be re-run by anyone, not just taken on the publisher's word. This mirrors how regulators overseeing structured digital reporting think about tagging quality: what matters is not that data carries tags, but that the tags are demonstrably correct.
What We Delivered
Three delivered examples of the same discipline. In each case validation is done with SHACL, the W3C's constraint-validation standard: a formal rulebook stating what a well-formed record must look like, checked by machine. Zero violations means every statement in the dataset passed.
| Delivery | Scale | Validation |
|---|---|---|
| Skills England Occupational Maps ontology | 51,355 triples | 0 SHACL violations |
| open-ontologies toolkit | Open-source validation primitives | Deterministic machine checks |
| Heritage aerial-photography archive standard | 292 digitised frames | 0 SHACL violations |
The national occupational maps, as a validated ontology
We converted the Skills England occupational maps into an open, formally structured ontology: 51,355 triples covering occupational standards and the relationships between them, validated at zero SHACL violations. Every statement in the graph passed machine validation against the published constraint rules. The ontology and an interactive explorer are available in our Skills England occupational maps write-up.
An open toolkit for ontology quality
The checks themselves are not bespoke one-offs. We maintain open-ontologies, an open-source toolkit of validation primitives for ontology quality. The design principle is that quality checking should be a deterministic machine check, not an LLM's opinion: a term exists in a vocabulary or it does not, a record satisfies a constraint or it does not. The project has attracted external contributors, including an engineer at Hyundai, which is what open validation tooling should do: outgrow its authors.
A computation-ready heritage archive
The same discipline applies well beyond skills data. For a heritage aerial-photography archive we modelled 292 digitised frames computation-ready, again validated at zero SHACL violations, turning a digitised collection into structured data that analysis can be run on directly. The full account is in our computation-ready aerial heritage write-up.
Outcome
Each publication carries its own evidence of quality: a formal schema, a machine-run validation report and the open tooling to reproduce it. A public body, a researcher or a regulator does not have to trust our claim that the data is well-formed; they can re-run the checks and see the zero-violations result for themselves. That is the standard we believe all published structured data should meet, and it is the standard we hold our own work to.
"Anyone can publish tagged data. The question that matters is whether the tags survive machine validation. We publish the data, the rules and the checker together, so the answer is never a matter of opinion."
Fabio Rovai, Tesseract Academy
Open validation tooling
The open-ontologies toolkit, and the Skills England ontology it validates, are published for reuse.
