CUE as a Declarative Knowledge Graph Construction Language

Use Case Submission for the KG-Construct Community Group

Summary

apercue.ca uses CUE — a constraint language with lattice-based type semantics — to construct, validate, and serialize knowledge graphs entirely at compile time. Graph building, SHACL validation, and JSON-LD output happen in a single cue export invocation. No mapping language, no runtime pipeline, no triplestore.

This submission demonstrates CUE as a KG construction language and explains how it relates to the KG-Construct CG's work on declarative KG construction.

The Construction Pipeline (One Step)

Traditional KG construction requires a multi-stage pipeline:

Source Data → Mapping (R2RML/RML) → RDF Store → SHACL Validation → Serialization

CUE collapses this to:

Source Data (CUE structs) → cue export -e <projection>

The "mapping" is CUE type unification. The "validation" is CUE constraint resolution. The "serialization" is JSON-LD context injection. All three happen during evaluation — there is no separate step for any of them.

Example: Research Publication Pipeline

Five nodes define a publication pipeline. Each declares @type (what it is) and depends_on (what it needs):

"analysis-code": {
    name:       "analysis-code"
    "@type":    {Process: true}
    depends_on: {"sensor-dataset": true}
}

The #Graph pattern computes topology, depth, roots, leaves, ancestors, dependents, and impact sets from this declaration alone.

Evidence: SHACL Validation (computed)

A compliance rule requires that publications have upstream data. The #ComplianceCheck pattern produces a standard sh:ValidationReport:

{
    "@type": "sh:ValidationReport",
    "sh:conforms": true,
    "sh:result": []
}

This is produced during CUE evaluation. sh:conforms: true means the graph satisfies all constraints. If any resource violates a rule, unification with true produces bottom (_|_) and cue vet fails — the graph cannot be constructed in an invalid state.

Evidence: Provenance (computed)

The #ProvenanceTrace pattern maps dependency edges to PROV-O:

{
    "@type": "prov:Entity",
    "@id": "urn:resource:analysis-code",
    "dcterms:title": "analysis-code",
    "prov:wasAttributedTo": {
        "@id": "apercue:graph-engine"
    },
    "prov:wasDerivedFrom": [
        {
            "@id": "urn:resource:sensor-dataset"
        }
    ],
    "prov:wasGeneratedBy": {
        "@id": "apercue:graph-construction"
    }
}

Every resource becomes a prov:Entity. Dependency edges become prov:wasDerivedFrom links. Provenance is not annotated after the fact — it is structurally computed from the same graph that produces the KG.

Evidence: JSON-LD Context (computed)

The vocabulary registry produces a JSON-LD 1.1 @context:

{
    "@context": {
        "@base": "urn:resource:",
        "dcterms": "http://purl.org/dc/terms/",
        "prov": "http://www.w3.org/ns/prov#",
        "dcat": "http://www.w3.org/ns/dcat#",
        "sh": "http://www.w3.org/ns/shacl#",
        "skos": "http://www.w3.org/2004/02/skos/core#",
        "schema": "https://schema.org/",
        "time": "http://www.w3.org/2006/time#",
        "earl": "http://www.w3.org/ns/earl#",
        "odrl": "http://www.w3.org/ns/odrl/2/",
        "org": "http://www.w3.org/ns/org#",
        "cred": "https://www.w3.org/2018/credentials#",
        "as": "https://www.w3.org/ns/activitystreams#",
        "void": "http://rdfs.org/ns/void#",
        "dqv": "http://www.w3.org/ns/dqv#",
        "oa": "http://www.w3.org/ns/oa#",
        "rdfs": "http://www.w3.org/2000/01/rdf-schema#",
        "owl": "http://www.w3.org/2002/07/owl#",
        "xsd": "http://www.w3.org/2001/XMLSchema#",
        "apercue": "https://apercue.ca/vocab#",
        "charter": "https://apercue.ca/charter#",
        "name": "dcterms:title",
        "description": "dcterms:description",
        "depends_on": {
            "@id": "dcterms:requires",
            "@type": "@id"
        },
        "status": {
            "@id": "schema:actionStatus",
            "@type": "@id"
        },
        "tags": {
            "@id": "dcterms:subject",
            "@container": "@set"
        }
    }
}

Every field mapping (name → dcterms:title, depends_on → dcterms:requires) is a CUE constraint. Adding a vocabulary term means extending a CUE definition, not editing a context file.

Relevance to KG-Construct

CUE does not replace RML for arbitrary RDF construction from relational sources. It targets constrained domains where the schema is known and the graph structure maps directly to typed dependencies. For these domains, the entire KG construction pipeline reduces to a type-checked cue export.

Limitations

• Closed-world: all resources declared upfront
• DAGs only: no cyclic dependencies
• Not a general RDF toolkit: targets constrained dependency graphs
• Performance: pre-compute transitive closure for graphs exceeding ~40 nodes

References

• Core report — Full implementation evidence with 18 W3C specs
• github.com/quicue/apercue — Source (Apache 2.0)
• demo.quicue.ca — Interactive D3 graph explorer