Sybilproof transitive trust protocols

We study protocols to enable one user (the principal) to make potentially profitable but risky interactions with another user (the agent), in the absence of direct trust between the two parties. In such situations, it is possible to enable the interaction indirectly through a chain of credit or "trust" links. We introduce a model that provides insight into many disparate applications, including open currency systems, network trust aggregation systems, and manipulation-resistant recommender systems. Each party maintains a trust account for each other party. When a principal's trust balance for an agent is high enough to cover potential losses from a bad interaction, direct trust is sufficient to enable the interaction. Allowing indirect trust opens up more interaction opportunities, but also expands the strategy space of an attacker seeking to exploit the community for its own ends. We show that with indirect trust exchange protocols, some friction is unavoidable: any protocol that satisfies a natural strategic safety property that we call sum-sybilproofness can sometimes lead to a reduction in expected overall trust balances even on interactions that are profitable in expectation. Thus, for long-term growth of trust accounts, which are assets enabling risky but valuable interactions, it may be necessary to limit the use of indirect trust. We present the hedged-transitive protocol and show that it achieves the optimal rate of expected growth in trust accounts, among all protocols satisfying the sum-sybilproofness condition.