Separating Facts from Interpretations in Agent Knowledge Graphs

TL;DR Most KG-augmented LLM systems store observations and judgments in the same graph. This breaks down at scale: facts and interpretations have different lifecycles, different governance needs, and require different evolution mechanisms. I split them into two physical tables: Fact KG — objective