MEV-Boost: Merge Ready Flashbots Architecture
Executive summary
MEV-Boost is the operational realisation of proposer-builder separation (PBS), the most consequential infrastructural change to Ethereum block production since the Merge. The design is deceptively simple: validators (proposers) outsource block construction to a competitive market of specialised builders via a relay intermediary, accepting whichever builder offers the highest bid. The implementation is a sidecar binary that ships alongside the consensus client. The economic effect is profound — it transformed MEV from a centralising force concentrated in vertically-integrated mining operations into a competitive market where the proposer captures most of the surplus. By 2024 over 90% of Ethereum blocks were produced via MEV-Boost. The paper is more accurately a design specification with engineering notes than an academic paper, but its impact justifies long-form review. DI rates it nine on influence and eight on rigor.
Background and motivation
The MEV literature begins with Daian et al's 'Flash Boys 2.0' (2019), which documented priority-gas-auction front-running on Ethereum and named the resulting extraction 'miner extractable value'. By 2020-2021 MEV had grown into a multi-billion-dollar annual extraction with significant centralisation implications: vertically-integrated mining pools that combined block construction, MEV searching, and consensus participation could earn more per block than honest miners, creating incentive pressure toward concentration. Flashbots' early product, MEV-Geth (2021), allowed searchers to submit private bundles to a permissioned set of miners, professionalising MEV extraction but not addressing centralisation. The Merge to proof-of-stake was scheduled for September 2022, and the question of how MEV would interact with the new validator set was urgent: validators with home-staking setups (a key decentralisation goal) could not realistically run the sophisticated MEV-search infrastructure that professional builders were developing, which would create a competitive disadvantage that drove staking toward centralised pools. MEV-Boost was Flashbots' answer to this problem. The design was published on ethresear.ch in November 2021 under the title 'MEV-Boost: Merge ready Flashbots Architecture', was tested through the 2022 testnet merges, and the system was active from the moment of the Merge. The intellectual antecedent is Buterin's 'enshrined PBS' proposals from 2021, which envisaged splitting the proposer and builder roles at the protocol level. MEV-Boost is the off-protocol realisation of the same idea — same separation, different trust model. The paper-as-published is a combination of a design document, a relay specification, and an empirical analysis of timing games and trust assumptions; it is a less formal artefact than most academic papers but its operational influence is comparable.
Core technical contributions
The architecture has three roles. Searchers identify MEV opportunities and construct bundles (ordered transaction sequences) that capture them. Builders aggregate bundles plus public mempool transactions into full block proposals and submit them to relays with a bid (the value the builder is offering to pay the proposer). Relays receive submitted blocks, verify their validity, and present blinded headers to proposers. Proposers select the highest-bid header, sign it, and only then receive the full block contents from the relay. This last step — the blind commitment — is the core trust trick. The proposer commits to a block before seeing its contents, which prevents the proposer from stealing the MEV by extracting it themselves. The relay is the trusted intermediary that ensures the builder's block is valid, the bid is honoured, and the proposer cannot refuse the block after committing. The relay is trusted not to send invalid blocks (which would slash the proposer) and not to collude with builders to steal MEV; in practice multiple independent relays have operated (Flashbots, BloXroute, Ultrasound, Agnostic, Aestus, and others) and proposers can subscribe to multiple. The builder API is standardised so that any party can become a builder by running the open-source builder code; the relay is similarly open-source. Bid validity, payload availability, and relay timing are specified in the public documentation, which has the property that an alternative implementation can be written and deployed without coordination with Flashbots. The system supports timing-game analysis: relays must release the block to the proposer within a tight window before the proposer's slot deadline, and builders must submit bids early enough to be included. The design considers adversarial timing strategies and the network-latency conditions under which relays must reliably release blocks.
Methodology and rigor
The methodology is engineering-oriented rather than formal. The trust model is explicit: proposers trust the relay; the relay trusts itself to not collude with builders; builders trust the relay to honour their bids; and the protocol-level slashing rules constrain proposer misbehaviour. There is no cryptographic mechanism preventing a malicious relay from stealing MEV (by reconstructing the block or sharing its contents with a colluding builder), only economic and reputational constraints. The paper is candid about this: relays are 'trust-minimised' rather than 'trustless', and a future enshrined-PBS design at the consensus layer would remove the relay trust assumption. Where the analysis is more developed is the timing-game discussion: the design examines the adversarial scheduling problem and the timing-attack surface that relays must manage under network-latency constraints. The empirical analysis (post-launch, 2022-2024) is also strong: builder-market concentration metrics, proposer participation rates, MEV-revenue distribution, and validator-decentralisation outcomes are all measured and published in regular Flashbots transparency reports. By 2024 the empirical evidence supported the paper's central claim: MEV-Boost adoption exceeded 90% of slots, builder-market diversity was meaningful (top-five builders with ~70% market share, varying), and validator-pool MEV concentration had not materially worsened post-Merge. The methodology that is missing is a formal economic-game-theoretic analysis of the multi-relay equilibrium: what happens if one relay defects, how proposers should diversify across relays, and whether relay reputation can be gamed at scale. Subsequent literature (Buterin's various PBS posts, Drake's enshrined-PBS notes, academic work on commit-reveal block construction) has begun to fill this gap.
Strengths
The most important strength is that the design shipped and worked. MEV-Boost was operational from day one of the Merge in September 2022, and within six months had >90% slot adoption. The validator-decentralisation outcome — home stakers can earn comparable per-slot returns to large staking pools by running MEV-Boost — is the single most consequential outcome and validates the paper's central motivation. The competitive builder market has produced multiple distinct builder strategies (Flashbots, Beaverbuild, Rsync, Builder0x69, others), each pursuing different MEV-extraction approaches; the diversity of strategies is itself a form of decentralisation that vertical integration would have suppressed. The relay open-source model has been imitated outside the original Flashbots stack: BloXroute, Eden, Manifold, Aestus, Agnostic, and several others operate independent relays. The bid-headers transparency means MEV-Boost data is publicly observable, enabling academic and industry research on MEV markets that would not have been possible under the previous closed mining-pool regime. The system's modularity has enabled rapid evolution: optimistic relays (relays that release the block to the proposer before fully validating, to reduce latency) emerged in 2023; censorship-resistance variants emerged after the OFAC-compliance debate in late 2022; pre-confirmation services and inclusion lists have been integrated experimentally. The design has also been ported beyond Ethereum: BNB Chain adopted an explicitly PBS-style builder API (BEP-322, which drops the relay role), and Solana's Jito client implements an analogous out-of-protocol block-auction market.
Weaknesses and limitations
The trusted-relay model is the most important limitation. A malicious or compromised relay can steal MEV by sharing block contents with a colluding builder or by submitting an invalid block that gets the proposer slashed. The protocol's economic and reputational constraints have so far been adequate (no relay has been caught colluding with a builder to steal MEV through 2025, although in April 2023 a rogue proposer exploited a bug in the open-source mev-boost-relay code to unbundle blinded blocks and drain roughly $20m from sandwich bots — a software failure rather than relay collusion), but the trust assumption is real and the design cannot eliminate it without consensus changes. Censorship resistance was the second flashpoint: in late 2022 the major US-operated relays (Flashbots, BloXroute Regulated) began filtering OFAC-sanctioned transactions, and validators using only those relays effectively contributed to censoring transactions. The community responded with non-filtering relays (Agnostic, Ultrasound, BloXroute Max-Profit) and proposer-side filter-bypass mechanisms, but the episode demonstrated that off-protocol PBS creates a regulatory pressure point that protocol-level PBS would not. Builder-market concentration is a continuing concern: by 2024 the top three builders consistently produced 60-80% of MEV-Boost blocks, and the barrier to becoming a competitive builder (latency-sensitive infrastructure, exclusive orderflow deals) has grown. The paper acknowledged this risk but offered no design-level countermeasure. Exclusive orderflow deals — searchers committing flow to specific builders in exchange for revenue share — have raised concerns about a 'private orderflow' regime that undermines the public competitive market the paper envisioned. The paper does not analyse multi-block MEV (where strategies span multiple slots), inclusion-list mechanisms (later proposed to address censorship), or the interaction with rollup sequencers, all of which became active design questions after publication. Finally, the paper's economic analysis is light on long-run equilibrium: as MEV revenue grows, what does the proposer-builder revenue split converge to, and how does that interact with staking-yield economics?
Subsequent influence
The influence is operational and architectural. Operationally, MEV-Boost is the default block-production pipeline on Ethereum — alternative pipelines (vanilla local block construction) exist but are vestigial. Architecturally, the proposer-builder-relay separation has become the dominant frame for thinking about MEV economics: regulators, academics, application developers, and L2 designers all reason in PBS terms. The Ethereum roadmap has explicitly committed to enshrining PBS at the protocol level (ePBS), with multiple candidate designs (Two-Slot, Drake's MEV-Burn, EIP-7732 PTC) under active discussion through 2025. Layer-2 ecosystems have adopted PBS variants: Arbitrum's Timeboost (proposed 2023, activated on mainnet in 2025) is a sequencer-side bid-for-priority mechanism conceptually descended from MEV-Boost; Optimism's Sequencer Decentralisation roadmap incorporates similar ideas. Beyond Ethereum, Solana's Jito-Solana client (launched on mainnet in October 2022) is an MEV-Boost-equivalent for Solana validators, with even stronger empirical adoption (>80% of stake by late 2024). Non-EVM chains (Cosmos, NEAR) have begun integrating analogous designs. The academic literature on MEV is now a recognisable subfield with hundreds of papers per year, and MEV-Boost is a standard citation. Flashbots itself has scaled from a research collective to a substantial entity with multiple product lines (SUAVE, Protect, Block Builder, Rollup Builder), all of which trace conceptual lineage to the MEV-Boost design.
How it holds up in 2026
Three and a half years after launch, MEV-Boost is the canonical Ethereum block-production pipeline and its core architectural claims have been validated empirically. Validator decentralisation has held — home stakers earn competitive yields and the staking pool concentration that PBS critics feared has not materialised at the predicted rate. The competitive builder market is real and persistent, even if more concentrated than ideal. The relay-trust assumption has held against collusion — though the April 2023 mev-boost-relay bug exploit showed the software layer is attackable — and remains the design's most criticised property. Where the system has aged less well: censorship-resistance concerns have not been definitively resolved, exclusive-orderflow market structures have grown beyond what the paper anticipated, and builder-market concentration has trended upward. The Ethereum community's response — committing to enshrine PBS at the protocol level — is itself the strongest endorsement of the paper's core direction, but the specific MEV-Boost implementation will likely be superseded by ePBS in the next several hard forks. The paper's enduring legacy will be the conceptual move (proposer-builder separation as a design pattern) rather than the specific sidecar implementation. As an artefact that translated a research idea into operational infrastructure on a multi-trillion-dollar network within months, it is one of the most efficient research-to-deployment cycles in the crypto literature. DI's holistic view: the design was a pragmatic compromise that solved an urgent problem, and its compromises (trusted relays) have justified the protocol-level work that will eventually replace it. Foundational and operational simultaneously.
Strengths
- Operational PBS without consensus changes — shippable on Merge day
- Empirically validated validator decentralisation outcomes (>90% slot adoption)
- Open-source builder API enabling permissionless builder participation
- Multi-relay design provided redundancy and competition
- Modularity supported rapid evolution (optimistic relays, censorship-resistant variants)
- Direct conceptual influence on every subsequent L1 and L2 PBS design
Weaknesses
- Trusted-relay model cannot eliminate MEV-theft risk without consensus changes
- Censorship-resistance concerns from OFAC-filtering relay subset
- Builder-market concentration trended upward post-launch
- Exclusive-orderflow deals create private-market structures undermining public competition
- No formal multi-relay equilibrium analysis at publication
- Long-run revenue-split economics under-analysed
Key contributions
- Operational specification of proposer-builder separation (PBS) without consensus changes
- Trusted-relay intermediary design that lets validators commit to builder blocks before seeing contents
- Competitive builder market replacing miner block-construction monopoly
- Standardised builder-API enabling permissionless builder participation
- Empirical demonstration that PBS reduces validator centralisation pressure from MEV
- Open-source sidecar implementation immediately deployable post-Merge
- Bid-and-block separation with timing-game analysis for relay safety
TL;DR
Off-protocol PBS that shipped on Merge day and reshaped Ethereum block production. >90% slot adoption validated the validator-decentralisation thesis; trusted-relay compromise created the design pressure that motivates protocol-level ePBS.
Get DeFi Intel research in your inbox
Weekly long-form coverage of papers, incidents, jurisdictions, chains, tokens and the people building them. Free tier covers headlines; Pro adds the analyst-grade breakdowns.
Sources
- github.com/flashbots/mev-boost
- ethresear.ch — MEV-Boost: Merge ready Flashbots Architecture (Nov 2021)
- arxiv.org/abs/1904.05234 — Daian et al., Flash Boys 2.0 (2019)
- collective.flashbots.net — post mortem, April 3rd 2023 relay incident
- github.com/bnb-chain/BEPs — BEP-322 Builder API Specification
External references gathered from the body of this brief. Last reviewed 2026-07-15.