Related work

Two recent preprints address evidence and maintenance in agent-accessible knowledge graphs. The descriptions below are limited to their abstracts; their evaluations are not benchmarks of Quipu.

• Vishisht Choudhary, Keep Your Receipts: Trustworthy Cleanup of Agent-Built Knowledge Graphs (2026), studies the consequences of withdrawing source evidence. PolyGraph records alternative evidence bundles for derived artifacts and audits those receipts by re-derivation. Its results motivate checking downstream dependencies rather than assuming that deleting directly extracted facts repairs every affected answer. This complements Quipu’s temporal record and SHACL-governed reconciliation: a merge’s structural validity does not establish the fidelity of its evidence receipts.
• Akash Raj, Plant Memory: Technical Language Processing and a Verified Maintenance Ontology for Agent-Accessible Failure Knowledge from SAP PM Work Orders (2026), describes rule-first extraction, optional LLM review, reified facts with confidence, evidence and verification state, technician review and MCP access. Its reported evaluation uses a synthetic 400-work-order dataset. It provides an application-specific example of reviewed agent-accessible knowledge; Quipu’s store and merge work instead concerns a general RDF graph governed by explicit shapes. The abstract does not establish a comparison of their temporal or withdrawal semantics.

Official arXiv BibTeX: Keep Your Receipts and Plant Memory.