# PBS relay trust is the critical unresolved centralization point in out-of-protocol MEV-Boost architecture Proposer-Builder Separation (PBS) separates block construction from block proposal: specialized builders prepare maximally-profitable blocks, validators propose without needing to construct them. MEV-Boost implements this as an out-of-protocol sidecar. The architectural trust dependency is the relay: relays store blocks from builders in escrow, select the most profitable block for each validator, and release block contents only after the validator commits to proposing it. The relay's trust position is expansive. Relays can censor specific transactions or builders. They can withhold blocks to cause validator proposer slashing. They can selectively share block contents to give allied searchers ordering information. Unlike validators (slashable) and builders (competitive market), relays have no protocol-level accountability mechanism in the out-of-protocol design. Exclusive orderflow compounds the problem. When searchers route transactions exclusively to specific builders, those builders gain a structural information advantage over competitors. MEV-Share was designed to dissolve exclusive orderflow by letting searchers share transactions while receiving MEV rebates — but as long as exclusive flows exist, relay-builder combinations that control exclusive flows exercise disproportionate power. A theoretical collusion attack exists: a validator controlling 20% of stake with consecutive block proposals could reorg a builder's payload without being slashed. This requires approximately 6.7M ETH — large but not impossible for coordinated actors. EIP-7732 (enshrined PBS) proposes moving the PBS auction to the protocol level, resolving relay trust dependencies by making relay behavior protocol-enforceable. However, ePBS does not fully resolve MEV distribution fairness or censorship concerns — those require additional mechanisms. BuilderNet v1.2 (February 2025) introduced reproducible TDX image builds as an intermediate trust reduction step: relays can verify that builder software matches a known binary, reducing but not eliminating trust requirements. The trust model progression: MEV-Boost (relay trust) → BuilderNet with TEE attestation (reduced trust) → ePBS (in-protocol, minimal trust) → SUAVE (cross-chain standardized sequencing). --- Relevant Notes: - [[L2 sequencer centralization creates systemic liveness censorship and regulatory risks]] — relay centralization on L1 is the structural parallel to sequencer centralization on L2: both are single points of ordering control - [[cross-chain composability breaks security assumptions when protocols span multiple chains with different trust boundaries]] — cross-domain MEV creates demand for shared sequencing that amplifies relay centralization concerns Topics: - [[protocol-mechanics]]