Diviner: Agentic RCA with Human-Governed Adaptation for Large-Scale Infrastructure
arXiv:2610.21281v1 Announce Type: new Abstract: Root-cause analysis (RCA) in large-scale distributed systems is challenging because a single symptom can cross service, data, configuration, compute, and organizational boundaries. Responders must assemble incomplete evidence and coordinate experts who hold different parts of the causal path. Generic language-model agents do not solve this problem through tool access alone: they can use stale or mistimed context, interpret failed queries as absent…