A knowledge graph generated from proofs
Retrieved sources are stored hash-addressed and scored for trust; an LLM actor then generates the claim graph. Nothing is hand-edited, and the generating actor, model, assessment and proof all stay traceable.
One retrieval of one source is one observation. A proof, a trust score, a generating actor and a projection result belong to the observation, not to each claim — so observations are the unit here.
Run status
Recent observations
Checking status…