SPARQL & RDF interface¶

epistemic-graph is an RDF triple-store: you load RDF (OWL/RDFS or plain triples), the engine maps it onto the property graph, and SPARQL SELECT runs over it. RDF and the property graph are the same data — a triple is a node, an edge, or a property depending on its object.

Status snapshot: SPARQL SELECT/ASK/CONSTRUCT/DESCRIBE, UPDATE, true named graphs, and the W3C /sparql HTTP endpoint (feature sparql-http) are all supported. Content negotiation, rich FILTER, sub-SELECT, SERVICE federation, MINUS, negated property sets (!p), ORDER BY/VALUES/EXISTS, FROM/FROM NAMED, and an SPO/POS index with selectivity join-ordering are shipped (EG-050..057, EG-KG.ontology.order-by-values-exists/135). GeoSPARQL + RCC8/Egenhofer (EG-KG.ontology.concept-10/155), the full RDF serialization matrix (JSON-LD/TriG/N-Quads/RDF-XML, EG-KG.ontology.eg-concrete-syntax-matrix/137), SHACL/ShEx/ICV validation (EG-KG.ontology.concept-6/133/146), and OBDA/R2RML virtual graphs (EG-101) are also in. See the capability matrix.

RDF ↔ property-graph mapping¶

RDF construct	Maps to
IRI / blank-node subject or object	a graph node (id = canonical term string)
triple with a resource object `(s, p, o)`	a typed edge `s --p--> o`
triple with a literal object	a property on `s` (value + datatype + lang preserved)
`rdf:type`	folded into the node `type` property and kept as an explicit typing edge
named graph	a `GraphCore` in the registry with a `:NamedGraph` marker

Multi-valued literals for the same predicate (which a key-unique property map can't hold) go to an opt-in lossless quads redb table under the rdf-redb feature; without it, extras are counted in LoadReport.dropped_multivalue, never silently lost. Export round-trips to N-Triples by set-equality.

SPARQL SELECT¶

SELECT ?s ?o WHERE {
  ?s rdfs:subClassOf ?o .
  FILTER (?s != ?o)
}
ORDER BY ?s
LIMIT 100

Implemented algebra (compiled to scans over the property graph, not an embedded triple evaluator):

Patterns: BGP, property paths (^p, p/q, p|q, p+, p*, p?, negated !p — EG-KG.ontology.negated-property-set), OPTIONAL, UNION, MINUS (EG-KG.ontology.minus), sub-SELECT (EG-KG.ontology.sub-select), SERVICE federation (EG-KG.query.sparql-service-federation-client), VALUES inline data.
Solution modifiers: Project, Distinct, Reduced, Slice (LIMIT/OFFSET), ORDER BY (multi-key ASC/DESC with SPARQL term ordering, incl. top-k ORDER BY … LIMIT — EG-KG.ontology.order-by-values-exists/135), Group + aggregates (COUNT/SUM/AVG/MIN/MAX/GROUP_CONCAT/SAMPLE), BIND, FROM/FROM NAMED dataset scoping (EG-KG.ontology.from-from-named).
FILTER (rich — EG-KG.ontology.rich-filter/127): arithmetic, REGEX, IN, EXISTS/NOT EXISTS, STR/LANG/LANGMATCHES/ DATATYPE/BOUND, isIRI/isLiteral/isBlank/isNumeric, string funcs (CONTAINS/STRSTARTS/STRENDS/STRLEN/SUBSTR/UCASE/LCASE/CONCAT/STRBEFORE/STRAFTER/REPLACE), hashes (MD5/SHA*), date-time (NOW/YEAR…/TZ), numeric (RAND/ABS/ROUND/CEIL/FLOOR), term constructors (IRI/BNODE/STRDT/STRLANG/UUID), COALESCE, IF, with datatype-aware comparison.

Query execution uses hashed SPO/POS/PSO indexes over the snapshot with cardinality-based BGP reordering (EG-KG.ontology.capability-catalog) — no longer a full scan per triple pattern.

ASK returns a boolean; CONSTRUCT instantiates its template; DESCRIBE returns a concise bounded description. UPDATE (INSERT/DELETE DATA, DELETE/INSERT WHERE, CLEAR, CREATE/DROP GRAPH, and the graph-management COPY/MOVE/ADD — EG-134) mutates the graph through a GraphStore; LOAD is intentionally unsupported (the write path does no HTTP fetch).

Content negotiation & serialization matrix (EG-KG.ontology.content-negotiation-serializers/136/137)¶

/sparql is Accept-aware, with a per-form default and an output=/format= override:

SELECT/ASK → SPARQL-results JSON / XML / CSV / TSV.
CONSTRUCT/DESCRIBE → Turtle, N-Triples, and — completing RDF 1.1 — JSON-LD 1.1 (expansion/ compaction, @id/@type/@graph), TriG + N-Quads (named-graph-aware), and RDF/XML.

The same matrix is available on ingest (parse) and via the W3C SPARQL 1.1 Graph Store Protocol (GET/PUT/POST/DELETE on /rdf-graphs/…?graph=, EG-134) for direct RDF-tooling read/write.

GeoSPARQL — spatial SPARQL (EG-KG.ontology.concept-10/155, feature `geosparql`)¶

The geo:/geof: vocabulary, WKT/GML geometry literals, and topological FILTER functions lower onto the eg-geo spatial predicates (see gis):

Baseline (EG-KG.ontology.concept-10): geof:sfWithin/sfIntersects/distance.
RCC8 (EG-KG.ontology.concept-7): geof:rcc8eq/dc/ec/po/tpp/ntpp/tppi/ntppi.
Egenhofer (EG-KG.ontology.concept-7): geof:ehEquals/disjoint/meet/overlap/covers/coveredBy/inside/contains.

SELECT ?site WHERE {
  ?site geo:hasGeometry/geo:asWKT ?g .
  FILTER (geof:sfWithin(?g, "POLYGON((…))"^^geo:wktLiteral))
}

Validation — SHACL / ShEx / ICV¶

SHACL (EG-KG.ontology.concept-6, crate eg-shacl): node/property shapes, sh:targetClass/targetNode, cardinality/datatype/pattern/sh:in/logical + SPARQL-based constraints → an sh:ValidationReport via Method::ShaclValidate.
ShEx (EG-KG.compute.concept-2): shape-expression schema validation (shape refs, triple constraints, cardinality, node constraints, AND/OR/NOT) via Method::ShexValidate.
ICV (EG-KG.ontology.wired-into-commit-write): Stardog-style Integrity Constraint Validation — interpret SHACL shapes under the closed-world / unique-name assumption as database integrity constraints, report violations with a focus-node + failing-constraint + a SPARQL "explain" witness, with an optional guard mode that rejects a write/transaction that would violate a constraint (and also runs in the OWL-reasoned view).

OBDA / virtual graphs — R2RML (EG-101)¶

R2RML-style :TriplesMap mappings (logicalSource → subject/predicate/object templates) expose a foreign/relational source (via the EG-073 ForeignSource registry) as a virtual RDF graph: SPARQL over the virtual graph rewrites to an Op::ForeignScan against the backing source — no materialization.

How it's reached¶

SPARQL is served two ways: the W3C /sparql HTTP endpoint (src/server/sparql_http.rs, feature sparql-http — GET ?query=, POST application/sparql-query/application/sparql-update, content-negotiated output), and the binary RPC method (Method::Sparql) with result-cache + RLS row-filtering. Companion RDF methods: AddTriples (durable, replicated), GetRdf (serialization-matrix export), and the validation methods above. SERVICE fan-out obeys the EPISTEMIC_GRAPH_SPARQL_SERVICE_ALLOW SSRF allowlist (fail-closed by default).

Reasoning¶

Loaded ontologies are reasoned over by the OWL 2 EL⁺/RL engine — see the ontology lifecycle guide.