Modeling Explainable AI Agent Payment Decisions Using Knowledge Graphs: Tradeoffs and Approaches

Hi everyone, Credian team here. We are exploring how to make AI agent payment decisions explainable after the fact, and I would love a knowledge graph perspective. If you needed to reconstruct why an agent was allowed to make a payment, how would you model the relationship between the human who delegated authority, the agent, the active policy, the approval, and the final transaction? Would you treat each policy decision as an immutable event node, or derive it from versioned entities and relationships? Our sandbox is still under construction, so I am mainly looking for the modeling tradeoffs. More context is at credian.io.