Explainable AI fund agents: show receipts — lenses → thesis → on-chain actions. You can audit what drove the call (narrative, flows, TA/macro), track execution, and define invalidation before sizing risk. That’s why “glass-box first, autopilot later” matters: it lowers