Docs index
Use the ontology output
Import the Compiled World ontology into RDF, linked-data and graph systems without losing validation, evidence or stable identity.
The two projections in every package
Every portable package includes ontology/knowledge.jsonld and ontology/knowledge.ttl. They are semantic projections of the Compiled World. The current export is RDF/JSON-LD with TAVONEL node kinds and compiled relation predicates; it should not be described as a hand-authored OWL/TBox schema. Keep provenance and validation beside the ontology, because the ontology is for semantic navigation and integration rather than a replacement for source evidence.
The predicates each engine emits
| Predicate | Emitted by | How it is derived |
|---|---|---|
| supported_by | Live engine and fallback engine | A claim to the evidence for one exact document version, carrying the evidence id |
| mentions | Live engine only | A claim to an entity, from a case-folded match over capitalised phrases, acronyms and Korean organisation names in that claim's own sentence |
| contradicts | Live engine only | Two claims that disagree on a number or on whether something is the case, inside one topic and one time reference. A candidate for review, not a resolution, and blind to jurisdiction, units and exception clauses |
| mentions_entity | Fallback engine only | Document-level co-occurrence from a capitalised-token scan, not read semantics |
| discusses_topic | Fallback engine only | Document-level co-occurrence from a small keyword rule set; the live engine has no Topic kind |
Note Relations the ontology vocabulary could express and no engine emits — supports, supersedes, depends_on — are not in a package. Query for a predicate that is not in the table above and the result is empty rather than wrong. contradicts is in a package now and was not before, so a query written against an older export finds nothing rather than nothing being there; and a contradicts row is a flagged pair awaiting a person, never a decided conflict.
Loading it into a target system
| Target | Use | Important companion |
|---|---|---|
| JSON-LD / linked-data application | ontology/knowledge.jsonld | provenance/activities.jsonl + validation/report.json |
| RDF store / SPARQL / RDF tooling | ontology/knowledge.ttl | canonical/model.json + provenance/ + validation/ |
| Neo4j or another property-graph import | graph/nodes.csv + graph/relationships.csv | Keep the same stable ids and evidence ids |
| Production AI or agent | Prefer live MCP/API | The active World stays current and preserves access/evidence resolution |
- Read validation/report.json first. A review_required candidate is not approved organizational truth.
- Choose JSON-LD for linked-data JSON consumers, Turtle for RDF/SPARQL consumers, or the CSV graph for property-graph import.
- Preserve every urn:tavonel:<id> identifier. Those ids are the join key across ontology, canonical model, graph and evidence-bearing records.
- Do semantic traversal in the ontology, but resolve factual claims back to provenance/evidence before presenting them as grounded answers.
- Treat a signed ZIP as a snapshot. When the active World changes, import the newer signed projection or use MCP/API instead of editing the old snapshot in place.
Inspecting it with SPARQL
SELECT ?resource ?type ?label
WHERE {
?resource a ?type .
OPTIONAL { ?resource <http://www.w3.org/2000/01/rdf-schema#label> ?label }
}
LIMIT 50Note The Turtle projection serializes node types, rdfs:label values and compiled relations. The JSON-LD projection also carries node evidence references through the PROV mapping in its context. For exact source locators, keep the provenance/evidence material from the same signed package.
API version 2026-09-02.1 · reviewed 11 September 2026
Something here out of date or wrong? Report an issue with this page.