A line without a meaning is ambiguous
A network can connect a customer, contract and project, but the line itself may conceal several different relationships. The customer could own the project, fund it or simply be mentioned in a related document. Graph retrieval surveys discuss the role of structured relationships, while provenance research emphasizes explaining their origins. A useful organizational graph therefore needs relationship types that correspond to inspectable evidence rather than a generic connected-to label that invites users to infer too much.
Choose a small, explicit vocabulary
An ABOLTUS experiment could start with relationships such as cites, amends, concerns and supports. A contract might concern a project while an amendment modifies a specific contract version. Each relationship should have a source passage or a clearly labeled inference rule. Keeping the initial vocabulary small helps reviewers understand its meaning and find inconsistent uses. Expanding it should follow demonstrated needs, because every additional relation introduces extraction, validation and presentation requirements that can make the graph harder to trust.
Test relation confusion
A practical evaluation would present similar document pairs with different relationships: a draft referencing a contract, a signed amendment changing it and an email merely discussing it. Reviewers should check whether the edge label matches the evidence. A graph with fewer accurate relationships may be more useful than a dense network of vague associations. The design goal is not visual connectivity for its own sake, but a vocabulary that lets users explain exactly why two organizational objects are linked and what that link does or does not establish.