🔥 MEV Internals

How the post-Merge MEV pipeline works

Every Ethereum block now passes through a four-step supply chain: users submit transactions to the public mempool or a private order-flow service, searchers scan for profitable reorderings and package them into bundles, builders assemble bundles plus ordinary mempool transactions into the highest-value possible block, and validators (proposers) pick the winning block from a relay auction via MEV-Boost. Before The Merge, this whole pipeline collapsed into the miner — the miner was searcher, builder, and proposer. After The Merge, proposer-builder separation split those roles, which is why roughly 90% of Ethereum slots as of April 2026 are built by professional builder firms rather than by the validator themselves.

The economic lever that makes the pipeline work is the builder-market auction. Builders compete for the right to pay the proposer: beaverbuild, rsync, Titan, and Flashbots run continuous optimization processes that order thousands of candidate transactions, simulate the resulting blocks, and submit increasing bids to relays. The relay validates each block and forwards only the highest-bid sealed header to the validator. The validator signs the blinded header without seeing the contents, the relay reveals the body, and the block is gossiped. Because the validator never knew what was in the block until after committing, the builder can pay full value for the MEV it extracted without worrying that the proposer will steal the bundle.

At the edges of the pipeline, two designs are reshaping MEV capture. On the sender side, intent-based aggregators (CoW Swap, UniswapX, 1inch Fusion) ask the user to sign a preference — "swap 10 ETH for at least X USDC by time T" — and have competing solvers bid to fill it off-chain or via private settlement. This keeps the user's trade out of the public mempool entirely, neutralizing sandwich attacks while letting solvers still capture legitimate cross-venue arbitrage. On the protocol side, enshrined PBS (ePBS) is the Ethereum consensus roadmap's answer to the trusted-relay problem: move the commit-and-reveal step into the consensus layer itself so a proposer can auction its slot without a relay middleman.

Key concepts

Proposer-Builder Separation (PBS)

The post-Merge architecture that splits "who builds the block" from "who signs and proposes it." Builders run optimization engines, bid for the proposer's slot, and pay the proposer directly on-chain at block construction time. Out-of-protocol PBS today uses trusted relays (MEV-Boost); enshrined PBS (ePBS) is the in-progress Ethereum upgrade that moves relay duties into the consensus layer and eliminates the trusted middleman via a builder-attestation step.

MEV-Boost relay flow

A validator runs the MEV-Boost sidecar, which at proposal time queries registered relays (Flashbots, Ultra Sound, bloXroute, Agnostic, Aestus). Each relay forwards the highest bid from connected builders. The validator signs the blinded header, the relay reveals the full block, and the payload is gossiped. Five-plus relays hold nontrivial market share, with Flashbots and Ultra Sound splitting the majority of non-censoring flow and Agnostic specializing in OFAC-uncensored blocks.

Sandwich attacks and JIT liquidity

A sandwich front-runs and back-runs a victim swap to extract the victim's slippage tolerance; JIT liquidity on Uniswap V3 front-runs a large LP-targeted swap with a tight range and then withdraws, capturing the LP fees that would otherwise accrue to passive range LPs. Both are classic toxic-MEV patterns that intent-based aggregators (CoW Swap, UniswapX) neutralize by never publishing the swap to the public mempool.

Flash loans and atomic arbitrage

Flash loans from Aave, Balancer, or Morpho let a searcher borrow millions of dollars with zero collateral inside a single transaction, repaying plus a 0–0.05% fee before the bundle reverts. This is what makes cross-DEX arbitrage an inventory-free business — the searcher never needs to warehouse capital on-chain. It also enabled the canonical flash-loan oracle manipulation attacks (bZx, Harvest, Mango), which is why every serious protocol now reads a manipulation-resistant TWAP or a delayed oracle rather than a spot price.

Private order flow and OFAC censorship

Private RPC services (MEV Blocker, Flashbots Protect, BloXroute Protected) route a user's transaction directly to searchers or builders instead of the public mempool, eliminating sandwich exposure. The cost is that private flow is opaque: in 2022 the U.S. OFAC sanctions on Tornado Cash pushed several MEV-Boost relays to filter sanctioned transactions, and the MEV-Boost market bifurcated into "censoring" and "non-censoring" relays, with Agnostic and Ultra Sound (in its non-filtering mode) serving the latter market.

Time-bandit attacks and consensus-layer MEV

A time-bandit attack re-orgs a past block to steal MEV that was captured by the original proposer — theoretically possible on any chain whose finality is probabilistic. On post-Merge Ethereum, finality arrives every 12.8 minutes (two epochs), so re-orgs past finality are economically infeasible for attackers without a supermajority of stake. On smaller PoS chains and pre-Merge PoW systems the attack has been demonstrated and is one of the arguments for single-slot finality and fast-finality gadgets.

Why MEV matters

As of April 2026, cumulative MEV extraction on Ethereum post-Merge is approaching $1.5B, and MEV-Boost secures somewhere between 85% and 90% of slots. Every Ethereum user pays MEV indirectly — through sandwich extraction on unprotected swaps, through slightly worse fills on public AMMs, through the gas auction that funds the builder bid that funds the validator — and every Ethereum-denominated staking return includes an MEV component that varies with volatility. A solo staker who runs vanilla self-built blocks leaves roughly 0.5–1.5% of annualized yield on the table; a validator who runs MEV-Boost captures it but pays a trust-assumption on the relay. That tradeoff is the core of the current ePBS debate.

MEV is also the clearest example in DeFi of how a market externalizes costs onto users by default and then gradually builds infrastructure to reinternalize them. Sandwich attacks were essentially free to run in 2020, gated only by mempool visibility; by 2026 they are neutralized on most sophisticated swap flow because CoW Swap, UniswapX, 1inch Fusion, and MEV Blocker all route user intent through protected channels. Liquidations still happen publicly because the externality is positive — a fast liquidator is what keeps lending protocols solvent — but the builder-market auction ensures the liquidation fee flows to whoever can execute the liquidation cheapest rather than to whoever controls the mempool. Understanding MEV-Boost, the relay flow, flash loans, and intent-based protection is therefore the frame for understanding how Ethereum economics actually route value after the Merge.

Frequently asked questions

What is MEV-Boost and why did 90% of Ethereum validators adopt it within a year?

MEV-Boost is the out-of-protocol implementation of proposer-builder separation. A validator runs the open-source MEV-Boost sidecar next to its consensus client; when it is selected as proposer for a slot, the sidecar polls a set of relays, each forwarding bids from competing block builders. The validator signs a blinded header of the highest bid, the relay reveals the full block body, and the block is gossiped. Adoption hit ~90% of slots within a year of The Merge because a solo-staker running a vanilla self-built block captures only the base-layer reward (~2–3% APR), while MEV-Boost lifts realized APR by another ~0.5–1.5% depending on volatility. The choice is not ideological — it is a direct cash yield.

How does a sandwich attack extract value and what does it cost the victim?

A searcher monitors the public mempool for a large AMM swap with non-trivial slippage tolerance. It then submits a bundle containing three transactions: a front-run swap in the same direction to push the price, the victim's swap executing at the worse price, and a back-run swap reversing the searcher's position. The victim sees a fill at the edge of their slippage tolerance while the searcher pockets the price impact minus gas. Typical extraction is 15–40% of the victim's declared slippage; an Eigenphi dashboard snapshot in 2025 showed median sandwiches netting $20–$50 on $5–15k Uniswap V2 swaps, and the top decile of multi-pool sandwiches clearing $1,000+ per opportunity. Setting slippage tighter than 0.5% and routing through a protected RPC (MEV Blocker, Flashbots Protect) defeats the attack.

What is the difference between MEV-Boost, MEV-Share, and ePBS?

MEV-Boost is the original relay-mediated auction: searchers send bundles to builders, builders send sealed blocks to relays, relays forward the highest bid to proposers. MEV-Share (the Flashbots user-protection layer) lets users submit transactions with privacy preferences and gives them a rebate when a searcher includes the transaction in a profitable bundle. ePBS is enshrined proposer-builder separation — the Ethereum roadmap proposal to move relay functionality into the consensus layer, eliminating the trusted relay as a middleman. ePBS adds an execution-payload-header slot and a builder attestation, so a proposer can commit to a builder's block without trusting an off-chain relay to honor the reveal.

How is the MEV-Boost relay market structured and who runs relays?

A relay sits between builders and proposers. It validates the builder's block (no invalid txns, no exceeding gas, no MEV theft), computes the effective bid, and forwards the blinded header to the proposer. The five largest relays by market share are Flashbots, Ultra Sound Relay, bloXroute, Agnostic (censorship-resistant), and Aestus. They are legally trusted counterparties — a relay that equivocates can rug proposers by showing one header and revealing another block, which is why ePBS exists. Builders compete on the relay market (rsync-builder, beaverbuild, Titan Builder, Flashbots, EigenPhi's Penguin Build), and the relay publishes proposer payments and builder bids in near-real time so the whole auction is auditable.

What is a flash loan's role in MEV and why do arbitrage bots use them?

A flash loan on Aave, Balancer, or Morpho borrows arbitrary principal inside a single transaction against no collateral, reverting the entire bundle if the principal plus fee (0.05% on Aave V3, 0 on Balancer) is not repaid by the end of the call. For arbitrage that means a searcher does not need to keep inventory on-chain: a $5M cross-DEX arb between Uniswap and Curve can be opened, executed, and closed with zero capital at risk, because the flash loan pays for itself from the spread. This is why atomic arbitrage bot flow is dominated by a handful of contracts that never hold inventory — the ROI on keeping capital idle lost to the ROI on borrowing it for one block.

What counts as toxic MEV vs beneficial MEV?

Toxic MEV is any extraction that makes a user's trade strictly worse than it would have been in an adversary-free execution: sandwich attacks, JIT liquidity attacks on Uniswap V3 LPs, time-bandit re-orgs. Beneficial MEV is extraction that improves market quality: cross-DEX arbitrage (keeps prices aligned, rewards accurate market-making), liquidations (clears bad debt, protects lender solvency), back-running to rebalance stale oracle prices. The split matters for protocol design — private RPCs and intent-based aggregators (CoW Swap, UniswapX, 1inch Fusion) explicitly aim to let users capture arbitrage MEV while blocking sandwich MEV by never publishing the user's intent to the public mempool.

How much MEV has been extracted on Ethereum since The Merge?

As of April 2026, cumulative MEV extraction on Ethereum post-Merge is approaching $1.5B as measured by EigenPhi, flashbots-data, and zeromev, with arbitrage (55–65% of volume), liquidations (15–25%), and sandwich attacks (10–15%) making up the bulk. Weekly extraction averages $3–8M in calm markets and spikes above $40M during ETH implied-volatility events. Roughly 85–90% of slots are built via MEV-Boost; the remaining 10–15% are self-built by solo-stakers who prefer vanilla blocks for censorship-resistance or ideological reasons. Those numbers set the baseline revenue figure that ePBS, intent-based aggregators, and L2 sequencer-auction designs are all negotiating over.

Related reading

• Sandwich attacks — mempool-watch simulator, slippage-tolerance math, and the four-transaction pattern that bots use to extract fills.
• Arbitrage mechanics — cross-DEX spread scanning, atomic Uniswap-vs-Curve trades, and the flash-loan pattern that lets searchers run zero-inventory arb.
• Flashbots and MEV-Boost — step through the relay flow, the blinded-header signature, and the builder market share across Flashbots, Ultra Sound, and Agnostic.
• Proposer-builder separation — compare out-of-protocol PBS (MEV-Boost) with enshrined PBS (ePBS) and see why the builder market looks the way it does.
• Aave liquidations — how liquidation MEV interacts with health-factor math, close-factor caps, and the 5–10% liquidation bonus that compensates keepers.