MEV-Boost is the middleware that implements proposer-builder separation on Ethereum: validators outsource block construction to specialized builders through relays, accept the most profitable header they see, and sign it blind. Since 2022 it has produced the large majority of Ethereum blocks, making it the single most consequential piece of tolerated infrastructure on the network. How it works The flow separates "who proposes" from "who builds": Builders assemble: sophisticated builders (often affiliated with professional searchers) construct MEV-optimized blocks — sandwich-arbitrage packed, prioritized txs ordered for profit. Relays relay: relays fetch builder blocks, validate them, and show the proposer only the header (commitment) — a design that prevents a builder from reneging on the proposer's payment ("header-only auctions"). Validators choose: the validator runs MEV-Boost, compares bids across relays, signs the most profitable header, and discovers the block contents only after signing — protecting the proposer from being tricked into rewarding an attesting thief. Settle: payment flows on-block to the proposer; builders and relays earn in the margins of MEV they captured. Because this happens at the middleware layer, it required no consensus change: MEV-Boost sits between the validator client and the network, and every participant can opt out at any time. Why it matters for decentralization MEV-Boost is a double-edged upgrade to Ethereum's economics. On the upside it democratized MEV revenue: solo validators earn the same blockspace upside as industrial operations without writing their own search bots. On the downside it concentrated MEV power: a handful of builders and relays now construct the overwhelming majority of blocks, which is a new form of infrastructural gatekeeping that the network never formally voted on. The censorship wake-up call was real: when OFAC-compliant relays captured a dominant share of block flow during sanction enforcement, the network faced de facto censorship of US-sanctioned transactions at the relay layer — exactly the "neutral infrastructure that quietly isn't" failure the software and governance pillars of our scoring model are built to detect. The Ethereum audit counts relay and builder diversity as first-class signals for this reason. Example: relay concentration on Ethereum today Check any MEV-boost dashboard (Titan, Flashbots, MEVWatch) and you will see the picture: a small set of relays captures the plurality of built slots, and the top few builders construct most blocks. The healthy property is that you can see this — relay market share, builder market share, and censorship attempts are all public data, refreshed per epoch. The fix being built toward it is enshrined PBS (protocol-level builder-proposer separation), which removes trusted relays from the loop entirely, plus encrypted mempools that starve builders of the information they need to sandwich you. Risks & limitations Relay oligopoly: if a few relays dominate, they become de facto ordering censors — measurably, as the OFAC episode proved. Builder concentration: even with diverse proposers, a small builder set centralizes where blocks get assembled and what order they propose. Timing games: relay latency and bid timing create subtle advantages for the fastest infrastructure, slowly pricing solo validators out of the top of the bid ladder. Infrastructure debt: "neutral" middleware that everyone relies on keeps needing patches against its own economic incentives — the pattern Watts, Titan, and Flashbots fight daily. What it looks like on a real slot Worked example, one beacon-chain slot: top builders bid around ~0.04–0.2 ETH depending on mempool conditions; a proposer running MEV-Boost sees the highest header and signs it; the builder then reveals full block contents (which include the sandwich and the arbitrage bundles) and its payment arrives on-block. Compare that to a proposer without MEV-Boost, which constructs purely by its own view of the mempool and typically misses the arbitrage it never saw. The numbers compound: historical data shows MEV-Boost proposers collecting a majority of the network's MEV revenue while non-participants forgo most of it. The MEV-Boost relay table, in practice RelayIn their serviceQuirk to know FlashbotsLargest historical share; research originOFAC-filtered; share dropped after impropriety reviews TitanFastest bids historically; strong proposer rewardsCentralization criticism common Ultra SoundNoeth-builder-adjacent; varietyMarket share is small but core diversity AestusIndependent, non-filteringSmall, but censorship-resistant signal Total relay-measurability is the whole point: any researcher can rebuild the market-share table from public slot data, and diversity scores can be tracked over time — the same discipline our chain audits use for every pillar. Why the variation across relays matters Two proposers querying the same set of relays can see materially different best-bids, because relay latency and builder-call quality vary continuously. That is not noise — it is the mechanism that produces relay market concentration in slow motion: proposers run, for example, three relays (a latency winner, a diversity winner, and a compliance-exception), and revenue-tracking rewards the fastest subsets. The pattern is identical to L2 sequencing: a small lead in latency compounds into majority flow, and the padding of "redundant" relays exists precisely to cap how far that compounding goes. Keep an eye on the in-protocol direction: with PBS making the builder/relay separation an official protocol feature, the trusted-relay-court design becomes the candidate that future MEV-Boost versions will progressively retire. And the practical floor: a proposer clipping income by half because it skimped on relays is a proposer subsidizing everyone else's order flow; on the relay side, proposer revenue outcome is the stat the ecosystem tracks monthly, and the fix remains the same diversified header set. Frequently asked questions Must validators use MEV-Boost? No — it is fully opt-in middleware. But opting out means leaving a substantial share of the block's value on the table, which is why adoption is near-total despite being voluntary. Does MEV-Boost hurt decentralization? It centralizes block building while keeping proposing decentralized — better than integrated builder-proposers, worse than the ideal. The question is whether relay/builder diversity keeps improving; the data says it is mixed. What actually fixes the risk? Enshrined PBS at the protocol level (moving the separation into consensus) plus encrypted mempool designs that remove the information advantage that builds the builder moat. Sources & methodology Flashbots MEV-Boost docs — the middleware and relay auctions. MEV-Boost relay/builder metrics — live market share per relay and builder. W3D methodology + academy dataset. Related terms MEV · Proposer-builder separation · Validator · Front-running · Sandwich attack Chain audits: Ethereum · Blast · tool: Validator profit calculator
MEV-Boost is the middleware that implements proposer-builder separation on Ethereum: validators outsource block construction to specialized builders through relays, accept the most profitable header they see, and sign it blind. Since 2022 it has produced the large majority of Ethereum blocks, making it the single most consequential piece of tolerated infrastructure on the network.
How it works
The flow separates “who proposes” from “who builds”:
- Builders assemble: sophisticated builders (often affiliated with professional searchers) construct MEV-optimized blocks — sandwich-arbitrage packed, prioritized txs ordered for profit.
- Relays relay: relays fetch builder blocks, validate them, and show the proposer only the header (commitment) — a design that prevents a builder from reneging on the proposer’s payment (“header-only auctions”).
- Validators choose: the validator runs MEV-Boost, compares bids across relays, signs the most profitable header, and discovers the block contents only after signing — protecting the proposer from being tricked into rewarding an attesting thief.
- Settle: payment flows on-block to the proposer; builders and relays earn in the margins of MEV they captured.
Because this happens at the middleware layer, it required no consensus change: MEV-Boost sits between the validator client and the network, and every participant can opt out at any time.
Why it matters for decentralization
MEV-Boost is a double-edged upgrade to Ethereum’s economics. On the upside it democratized MEV revenue: solo validators earn the same blockspace upside as industrial operations without writing their own search bots. On the downside it concentrated MEV power: a handful of builders and relays now construct the overwhelming majority of blocks, which is a new form of infrastructural gatekeeping that the network never formally voted on.
The censorship wake-up call was real: when OFAC-compliant relays captured a dominant share of block flow during sanction enforcement, the network faced de facto censorship of US-sanctioned transactions at the relay layer — exactly the “neutral infrastructure that quietly isn’t” failure the software and governance pillars of our scoring model are built to detect. The Ethereum audit counts relay and builder diversity as first-class signals for this reason.
Example: relay concentration on Ethereum today
Check any MEV-boost dashboard (Titan, Flashbots, MEVWatch) and you will see the picture: a small set of relays captures the plurality of built slots, and the top few builders construct most blocks. The healthy property is that you can see this — relay market share, builder market share, and censorship attempts are all public data, refreshed per epoch. The fix being built toward it is enshrined PBS (protocol-level builder-proposer separation), which removes trusted relays from the loop entirely, plus encrypted mempools that starve builders of the information they need to sandwich you.
Risks & limitations
- Relay oligopoly: if a few relays dominate, they become de facto ordering censors — measurably, as the OFAC episode proved.
- Builder concentration: even with diverse proposers, a small builder set centralizes where blocks get assembled and what order they propose.
- Timing games: relay latency and bid timing create subtle advantages for the fastest infrastructure, slowly pricing solo validators out of the top of the bid ladder.
- Infrastructure debt: “neutral” middleware that everyone relies on keeps needing patches against its own economic incentives — the pattern Watts, Titan, and Flashbots fight daily.
What it looks like on a real slot
Worked example, one beacon-chain slot: top builders bid around ~0.04–0.2 ETH depending on mempool conditions; a proposer running MEV-Boost sees the highest header and signs it; the builder then reveals full block contents (which include the sandwich and the arbitrage bundles) and its payment arrives on-block. Compare that to a proposer without MEV-Boost, which constructs purely by its own view of the mempool and typically misses the arbitrage it never saw. The numbers compound: historical data shows MEV-Boost proposers collecting a majority of the network’s MEV revenue while non-participants forgo most of it.
The MEV-Boost relay table, in practice
| Relay | In their service | Quirk to know |
|---|---|---|
| Flashbots | Largest historical share; research origin | OFAC-filtered; share dropped after impropriety reviews |
| Titan | Fastest bids historically; strong proposer rewards | Centralization criticism common |
| Ultra Sound | Noeth-builder-adjacent; variety | Market share is small but core diversity |
| Aestus | Independent, non-filtering | Small, but censorship-resistant signal |
Total relay-measurability is the whole point: any researcher can rebuild the market-share table from public slot data, and diversity scores can be tracked over time — the same discipline our chain audits use for every pillar.
Why the variation across relays matters
Two proposers querying the same set of relays can see materially different best-bids, because relay latency and builder-call quality vary continuously. That is not noise — it is the mechanism that produces relay market concentration in slow motion: proposers run, for example, three relays (a latency winner, a diversity winner, and a compliance-exception), and revenue-tracking rewards the fastest subsets. The pattern is identical to L2 sequencing: a small lead in latency compounds into majority flow, and the padding of “redundant” relays exists precisely to cap how far that compounding goes. Keep an eye on the in-protocol direction: with PBS making the builder/relay separation an official protocol feature, the trusted-relay-court design becomes the candidate that future MEV-Boost versions will progressively retire.
And the practical floor: a proposer clipping income by half because it skimped on relays is a proposer subsidizing everyone else’s order flow; on the relay side, proposer revenue outcome is the stat the ecosystem tracks monthly, and the fix remains the same diversified header set.
Frequently asked questions
Must validators use MEV-Boost?
No — it is fully opt-in middleware. But opting out means leaving a substantial share of the block’s value on the table, which is why adoption is near-total despite being voluntary.
Does MEV-Boost hurt decentralization?
It centralizes block building while keeping proposing decentralized — better than integrated builder-proposers, worse than the ideal. The question is whether relay/builder diversity keeps improving; the data says it is mixed.
What actually fixes the risk?
Enshrined PBS at the protocol level (moving the separation into consensus) plus encrypted mempool designs that remove the information advantage that builds the builder moat.
Sources & methodology
- Flashbots MEV-Boost docs — the middleware and relay auctions.
- MEV-Boost relay/builder metrics — live market share per relay and builder.
- W3D methodology + academy dataset.
Related terms
MEV · Proposer-builder separation · Validator · Front-running · Sandwich attack
Chain audits: Ethereum · Blast · tool: Validator profit calculator