Quipu: A Governed Bitemporal Knowledge Graph Store
A recent publication on arXiv (2608.16813) presents Quipu, a knowledge-graph storage solution that aims to overcome the limitations of conventional systems when faced with agent workloads. The authors contend that existing stores are configured primarily for human curation, allowing immediate writes with later cleanup, using a single or no time axis, treating all writers' data as equally credible, and leaving governance to dashboards and middleware. Quipu challenges these norms by implementing a gate that evaluates facts before entry; employing bitemporal data, trust labels, and rules; utilizing named graphs as the basis for authority and trust within a lattice structure; and storing governance specifications, traces, and signed verdicts as facts, simplifying the audit query T ⊨ Σ. The system is assessed through Census, a deterministic multi-writer lifecycle. This paper is classified as a new announcement and can be accessed via the provided URL.
Key facts
- Quipu is an embeddable knowledge-graph store.
- It inverts four defaults of traditional knowledge-graph stores.
- No fact enters without passing through a gate with predicates.
- Data, trust labels, verdicts, and rules are bitemporal.
- Named graphs are the unit of authority and trust.
- Composition under a lattice never widens.
- Governance specification Σ, trace, and signed verdicts are facts in the store.
- Evaluation uses Census, a deterministic multi-writer lifecycle.
Entities
Institutions
- arXiv