Keep Your Receipts: Trustworthy Cleanup of Agent-Built Knowledge Graphs

Key points

• When a source is retracted, overruled, or erased, everything built on top of it (entity merges, entailed edges, community summaries) must be re-examined, and the maintainer must choose between deleting too much and too little.1
• Deployed pipelines delete only the directly extracted facts: on our corpus that reduces contaminated queries from 32.2% to 28.0% under a real retraction and changes nothing under predicate revocation.1
• Because receipts are the model's own self-report, a re-derivation audit checks them by deleting the cited evidence and testing whether the constructor still produces the artifact.1
• The audit finds that a receipt is trustworthy when it is shaped like the claim it backs.1

Sources (1)

• [1]Keep Your Receipts: Trustworthy Cleanup of Agent-Built Knowledge Graphs
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 13, 04:51 AM

  Agents build knowledge graphs as memory, and building them has become easy.

  When a source is retracted, overruled, or erased, everything built on top of it (entity merges, entailed edges, community summaries) must be re-examined, and the maintainer must choose between deleting too much and too little.

Extractive summary: sentences quoted from the sources.