DIG to Heal: Explainable Dynamic Decision Paths Scale General-Purpose Agent Collaboration

arXiv·medium signalXBlueskyLinkedInCopy link

arXiv 2603.00309 proposes Dynamic Interaction Graph (DIG) to Heal — a framework using explainable dynamic decision paths to coordinate multi-agent collaboration in general-purpose tasks, providing human-inspectable intermediate decision graphs. The approach addresses the opacity problem in multi-agent pipelines where agent-to-agent communication is opaque and interventions are difficult. Particularly relevant for compliance-sensitive deployments where audit trails and human-in-the-loop override points are required alongside autonomous agent execution.