Claim-Gated Source-Risk Auditing for Generative Search

The paper introduces a claim‑gated audit framework for generative search, ensuring that a query, source, and answer tuple is only considered resolved when relationship evidence, answer adoption, materiality, and disclosure are all present. It distinguishes this audit endpoint from citation support and review priority, tying decisions to versioned evidence spans and implementing a reference checker to enforce the contract. Experiments on a synthetic dataset confirm that the system correctly handles all 81 predicate combinations and rejects 192 malformed records, while ablation studies isolate endpoint logic from missing‑evidence handling.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv AI.