Semogram Docs
OntologyReference

Terms and contracts

Compare model, binding, query, fact and claim structures

Identity and scope

ResourceScopeVersion identifier
Definition assetProjectSaved-version UUID; active-version selection
Compiled ontology packageProjectPackage UUID plus numeric package version
Read bindingProject; selects workspace endpointNumeric binding version/status
Query authoring recordProjectSaved query-version UUID
Query releaseProject/queryRelease number and release ID
Materialization pipelineProjectSaved graph version and run snapshot
Assertion / claimWorkspace and project in these workflowsClaim revision and decision history
Evidence captureWorkspace/projectPreserved acquisition/object integrity

Definition input

Definition creation uses manifest and files. Manifest includes name, description, defaultNamespace, modelEntrypoints, shapeEntrypoints, imports and optional definitionImports. Each file has a relative path, content and optional contentType. Commit message is optional. Updates commit a full package; omitted files are not an instruction to retain an old file implicitly.

Binding fields

FieldRequirement/meaning
ontologyPackageId/versionCompiled identity and positive numeric version
termRef/termIriExact model identifier; use the full IRI in these examples
termKindclass, property, relationship, identity or wildcard
kindmaterialized, virtual, asserted or inferred
dataEndpointIdCompatible accessible workspace endpoint
statusdraft, active or archived
cardinalityone, optional_one, many or unknown
priority/default/versionBinding selection metadata
resolutionMethod-specific source mapping/filter/expression
accessPolicySupported permission/role/masking restrictions

MCP create wraps these under binding; HTTP create accepts them as the body. The project comes from the HTTP route or authenticated MCP scope. Do not copy an HTTP body directly into MCP arguments without its wrapper.

Query format

Saved query creation accepts name, description, content (YAML string), authoringMode and optional aiPrompt. The YAML query_definition contains compiled package identity, target, root, schemas, output_mapping, optional filters/sort and policy. Publication uses expectedVersionId. Execution uses input and options, not the author's full YAML.

Input/output schemas use supported JSON Schema draft-07. Output schema describes each row. Execution options allow a page limit of 1–1000, offset, cursor or jobId, with continuation restrictions. Retained result jobs currently expire after one hour.

Facts and claims

An ontology mapping configuration uses packageId, inputs/output, mode, entities and relationships. Entity rules contain class IRI, stable ID expression, properties, source identities and provenance. Materialization selects package and fact-store endpoint, mode and enforcement. Node fragments belong inside a complete pipeline graph.

A basic human assertion uses subjectRef, predicate, objectRef or scalar value, assertionType, confidence and validity. Governed claims additionally distinguish origin, confidence dimensions, precise evidence links, revisions, decisions, verification and derivations. These contracts are not interchangeable. Pipeline extracted assertions require real evidenceLinks; a citation URL or free-form object is insufficient.

Important limits

Live, materialized, asserted and inferred values can coexist but must share an entity identity. Named-term entailment uses rdfs-owl-relations-v1, not complete OWL support. Consumer SPARQL is opt-in and bounded. Query release dependency checks can require republication after model/endpoint change. Generic public assertion mutation and verification-creation interfaces are not currently exposed.