Building a knowledge graph for real assets
How to connect documents, people, interests, properties, wells, payments, obligations, and geography without losing source truth.
Model identity before relationships
A useful graph begins with stable identities. Owner names, operator names, property labels, and legal descriptions vary across sources; the graph should preserve those aliases while resolving them to canonical records. Otherwise, every relationship inherits uncertainty that is hard to unwind later.
Keep facts, documents, and interpretations distinct
Structured operational facts belong in durable schemas. Documents remain evidence. Extracted or model-generated interpretations should link back to both, with confidence and provenance. This separation lets the platform improve extraction without silently rewriting authoritative history.
- Record where each fact came from and when it was effective.
- Represent ownership and assignments as time-bound relationships.
- Connect geography without replacing the legal description that created the right.
- Make unresolved matches first-class review work instead of hidden joins.
Design for questions and actions
The graph should support the questions people ask and the actions workflows take: which owners are affected, which document controls, what changed this month, what obligation is due, and which statement line supports the exception. Start from those operating questions rather than an abstract ontology exercise.