Why AI Agent Reputation Needs Both Link Analysis and Flow-Based Gating

Agent delegation chains now cross organizational boundaries. Two families of trust-inference algorithms are candidates for computing reputation in such networks, and each has known failure modes when deployed alone. Link-analysis algorithms such as EigenTrust compute reputation from the topology of pairwise attestations, but, as the original authors acknowledge, do not resist Sybil attacks on their own. Flow-based defenses such as SybilGuard, SybilLimit, and SumUp bound Sybil influence through attack-edge scarcity but do not produce a reputation ordering within the honest region. In an open agent network, both attacks are simultaneously realistic. This is a position paper. It argues that link analysis and flow-based gating are compositional rather than substitutable, gives a high-level formalization of the gate as a function of max-flow from a seed set, and describes a standards-based representation of the resulting reputation signal, expressed as a W3C Verifiable Credential with a Data Integrity Proof over a canonicalized subject, signed with Ed25519, and published through a did:key-resolvable signer endpoint. The paper does not claim novelty in either algorithm family individually; its contribution is the argument for composition in the agent-network setting and a concrete portability representation. A reference credential format and a live signer endpoint are documented, and a stable verification fixture is planned.