This could be just like discourse graph’s default ontology (types: questions, claims, evidence, etc. relationships: supports, opposes, informs, etc)

Alternatively, there could be a more general knowledge graph ontology provided (using existing work with RDF, for example).

Likely it will make sense to allow for multiple ontologies to exist as “translators” for users’ knowledge graphs

Side note: I’m thinking about category theory and considering knowledge graphs to be categories and translations between them to be functors.