Early Lessons from a Work in Progress
The Challenge
- Knowledge graphs represent what is true, but in domains like music and content classification, truth is contested, evolving, and subjective.
- When systems flatten ambiguity, they lose organizational intelligence.
- With AI agents now consuming the graphs, the problem is urgent.
- Agents can't distinguish institutionally-ratified claims from contested ones, producing not hallucination but false consensus.
- Emerging standards like OKF standardize a grain for structural metadata, but provide no vocabulary for epistemic metadata or typed relation: confidence, verification status, decision provenance, temporal validity.
A Solution
We propose a working augmentation: an approach that borrows both from OKF and RDF, but with separable conventions for the dimensions that emerge through collaboration at scale.
What we did
- What we built.
- The ambiguous situations that shaped each decision.
What we learned — including why inconsistency between sources is often signal, not noise.