Preprint
Aug 2026
Trace Integrity for LLM Data Agents: A Vision for Auditable Structured Reasoning in Real-World Systems
This paper introduces Trace Integrity, a deployment reliability criterion for evaluating whether the computation recorded behind an answer is explicit, executable, schema-valid, operator-faithful, replayable, answer-consistent, answer-consistent, and auditable.
Srimonti Dutta, Akshata Kishore Moharir
· 0 citations