arXiv:2606.19407v1 Announce Type: cross Abstract: Large language models can produce fluent root cause analyses, but fluent final answers alone are insufficient evidence for accountability in high-stakes operations. In real incident response, engineers need to know what evidence supported a diagnosis, which alternatives were considered, where contradictions remained, and whether the system resolved the case or preserved uncertainty. We address this gap with JustDiag, a diagnostic justification engine for RCA that maintains an explicit process state over evidence, findings, competing hypotheses,
Source: arXiv cs.AI — read the full report at the original publisher.