Ontologies, vocabularies and concepts actually used

Changes · Discovery method · Backlog

An ontology defines categories and relationships. A vocabulary can be a smaller set of terms. A namespace gives those terms stable identifiers. A declared namespace does not prove that its terms are used, that an ontology was imported, or that a reasoning engine applies it. This project uses a small set of established metadata and concept vocabularies, with explicit local terms. It does not import a complete benefits or legal ontology.

Implemented namespace map

This flowchart TD shows implementation responsibilities. Its arrows mean “uses these terms”; they are not additional benefit or legal assertions.

flowchart TD
    YAML[Authored YAML-LD and frozen source records] --> JSONLD[Generated JSON-LD and RDF]
    JSONLD --> Documents[Documents, pages and human tasks]
    JSONLD --> Concepts[Concepts and source-backed associations]
    JSONLD --> Assertions[Evidence-bearing relationship assertions]
    Documents --> SCHEMA[Schema.org: DigitalDocument, CreativeWork, Question, Role, text]
    Documents --> DCT[DCMI Terms: title, source, isPartOf, references, requires, licence]
    Documents --> DCAT[DCAT: keyword tags]
    Concepts --> SKOS[SKOS: Concept and related]
    Concepts --> DWP[Local dwp namespace: three experimental domain predicates]
    Assertions --> RDF[RDF: Statement, subject, predicate, object]
    Assertions --> OKF[Local okf namespace: status, scope, authority and evidence fields]
    Assertions --> PROV[PROV-O: derivation, primary source and observation time]
    Assertions --> RDFS[RDFS: readable labels]
    JSONLD --> XSD[XML Schema datatypes: dates, numbers and month precision]
    NAV[Additive navigation catalogue: 44 labels] --> SKOSNAV[SKOS ConceptScheme and Concept]
    NAV --> DISC[Literal discovery assignments with explicit unknowns]
    DISC --> BOUNDARY[No legal applicability or entitlement assertion]

Actual use and evidence

The pinned JSON-LD context, semantic compiler, generated pilot graph and full-DMG semantic manifest are the implementation evidence. The machine-readable namespace audit records the measured term use. Counts below describe the retained full-DMG semantic shards before the additive navigation catalogue; they are not completeness scores.

These are stored statements, not a materialised logical closure. Runtime records label 282 items “Concept”, while 268 have an explicit skos:Concept type in the retained RDF; some older records lack that type. Do not add the counts together or silently treat the difference as a second ontology. The CPAG Book is a metadata reference, not handbook-body evidence.

SKOS provides concept organisation and associations. This implementation's stored association does not establish legal equivalence. PROV-O supplies provenance terms; having provenance does not make an interpretation correct. Official vocabulary references were checked on 19 September 2026; local counts come from the frozen artefacts, not the vocabulary specifications.

Additive conceptual navigation

The new navigation catalogue contains 44 project-authored labels: 20 benefits, 10 circumstances and 14 topics. The navigation manifest accounts for all captured pages; literal matches and existing cited concept links are discovery evidence. An unknown classification remains explicit. These assignments do not mean a benefit rule applies to a person or that the page completely covers the topic. Existing concept IRIs are retained rather than rewritten.

Three different observations must stay distinct: the text contains literal X; an existing project concept cites page Y; and reviewed guidance applies to benefit Z. The navigation producer supplies the first two, with their evidence. It does not establish the third. It deliberately does not generate page schema:about, skos:related or appliesTo assertions from lexical hits: an example, footnote, exclusion or coincidental phrase may mention a benefit without making the whole page its guidance. The assignments remain governed runtime facets; the label catalogue uses SKOS separately.

Examples such as Benefit, BenefitVariant, Claimant, Circumstance, LegalCustody, Entitlement, Payment, Disqualification and DecisionRule are modelling questions. Their presence in requirements or prose does not mean they are formal OWL classes. The semantic pilot contract states the meaning and evidence requirements for the local predicates actually emitted. Neutral cross-benefit definitions and review are tracked as DWP-BL-005.

The staff and household increments add 51 source-backed concept proposals, including neutral benefit and benefit-variant meanings, using SKOS related and broader and Dublin Core references. It retains the earlier namespace audit unchanged. The 40 task profiles are executable context requirements, not an OWL benefits ontology. Their 203 missing obligations are explicitly absent evidence. Legal citation links identify provision references; reference metadata does not become statutory text or an applicability rule. Twenty selected statutory units now have separate derived evidence records and 43 source-backed references, using the same Dublin Core vocabulary. No additional legal ontology or applicability predicate is invented for them.

The qualification increments also use the existing Dublin Core requires term for 15 explicit support dependencies. The initial household checkpoint supplies eight: seven captured pages for the complete household summary, plus the care-home overview's dependency on that summary. The component follow-up adds five housing-cost and two temporary-residence page dependencies. Staff 012 and 013 require the household and housing-cost paths; temporary residence remains a record-level requirement. These are authored context needs with evidence and review status; they do not assert that a benefit is payable or that a legal rule applies. No new ontology predicate is introduced.

Declared or researched, but not implemented as claimed semantics

For another department, repeat discovery of its native terms and standards. Reuse the proven metadata stack where appropriate; do not copy local DWP predicates or benefit concepts as if they were universal.