MEV is the value someone can extract from a blockchain by controlling the contents and order of a block, over and above the ordinary fees and rewards for producing it. The acronym came from Flash Boys 2.0 (Daian, Goldfeder, Kell, Li, Zhao, Bentov, Breidenbach and Juels, April 2019), which documented arbitrage bots bidding transaction fees up against each other in what the authors called priority gas auctions, and argued that fees paid for ordering are large enough to be a consensus-layer security risk rather than a trading curiosity. It stood for “miner extractable value”; after Ethereum moved to proof of stake the expansion drifted to maximal extractable value, which is how CoW Protocol's documentation and most of the industry now write it.

The reason the concept refuses to go away is that ordering is a real economic right and somebody always holds it. A blockchain does not execute transactions as they arrive; it executes them in the sequence a block producer chooses. Whoever chooses that sequence can put their own trade first, insert one on either side of yours, or sell that privilege to someone who will. Every general-purpose chain has this property. The only open questions are who captures the value, whether the people who generate it get any of it back, and what the fight over it does to the rest of the system.

Where the value actually comes from

MEV is not a single trade. It is a family of strategies that share one input – a temporary, mechanical mispricing that only exists because state changes in a discrete order.

• DEX arbitrage. Two pools quote the same pair at different prices, or a pool's price drifts from the wider market. A bot buys on one side and sells on the other in a single atomic transaction. This is the benign end of the family: it is how AMM prices track reality at all.
• Liquidations. A lending position crosses its health threshold and the protocol pays a bonus to whoever closes it. The bonus is a fixed parameter; the race to claim it is a pure ordering contest.
• Sandwich attacks. A bot sees a pending swap large enough to move a pool, buys ahead of it, lets the victim execute at the worsened price, and sells behind it. CoW Protocol's documentation calls this “price exploitation” and notes it has cost ordinary traders more than a billion dollars. Unlike arbitrage, nothing about it improves the market – it is a transfer from the trader to the bot.
• Just-in-time liquidity. A bot adds concentrated liquidity in the same block as a large swap, collects the fees on that swap, and withdraws immediately, taking fee income from the passive providers who carried the inventory risk all week.
• CEX–DEX arbitrage. The largest and least visible category. Centralised venues reprice continuously; an AMM pool reprices only when someone trades against it. The gap is picked off every block by whoever is fastest.

That last category has a formal name. In Automated Market Making and Loss-Versus-Rebalancing (Milionis, Moallemi, Roughgarden and Zhang, 2022, revised 2024) it is called loss-versus-rebalancing, or LVR – presented as a “Black-Scholes formula for AMMs”. The paper identifies the adverse-selection cost an AMM liquidity provider bears because the pool quotes stale prices that better-informed arbitrageurs pick off, and derives closed-form expressions for it across all constant-function market makers. The important consequence for a DAO is that LVR is structural, not an attack: it is the standing cost of running an on-chain market maker against off-chain price discovery, and it shows up whether or not anyone behaves badly.

The supply chain that sells ordering

On Ethereum the ordering right is no longer exercised by the validator who proposes the block. It is auctioned through a four-layer pipeline that MEV-Boost formalised as proposer-builder separation:

1. Searchers find extractable opportunities and package them into transaction bundles with a bid attached.
2. Builders assemble whole blocks out of public mempool flow and searcher bundles, optimising for total extractable value, and quote a payment to the proposer.
3. Relays hold the built block, show the proposer only its bid, and release the contents once the proposer has signed – so neither side can steal from the other.
4. Proposers (validators) run MEV-Boost as a sidecar, take the highest bid, and sign a block they never saw.

The stated goal was decentralisation: separating who builds from who proposes was meant to let a home validator earn the same MEV income as a sophisticated one. On that narrow measure it worked. On concentration it did not. Over the seven days to 1 August 2026, relayscan.io recorded Titan building 53.61% of MEV-Boost blocks and Quasar 20.98% – two firms producing roughly three-quarters of them – with BuilderNet at 13.06% and Eureka at 8.81% making up most of the remainder. The relay layer is similarly thin: ultrasound.money (27.00%), bloXroute max-profit (23.29%), Titan Relay (20.10%) and bloXroute regulated (19.37%) carried about 90% of payloads between them. Block construction is a competitive market with a handful of competitors, which is a different thing from a decentralised one.

The proposed fix is to bring the auction inside the protocol. EIP-7732, “Enshrined Proposer-Builder Separation”, would make builders an in-protocol entity: the consensus proposer commits to a builder's signed bid, the value is deducted from the builder's own stake, and a payload timeliness committee attests to whether the builder actually revealed the block it promised. It was created in June 2024 and is still at Review status – a design under active discussion, not a shipped feature.

Who actually pays

MEV is often described as a tax on traders. That undercounts it. The bill lands on three groups, and only one of them is aware of it in the moment.

• Liquidity providers pay the largest and quietest share, through LVR. The fees look fine in the interface; the arbitrage losses net against them silently. Any DAO running protocol-owned liquidity, a treasury LP position, or a liquidity mining programme is paying this whether or not it is measured, and a mining programme that ignores it can be paying emissions to attract capital that loses money on arrival.
• Traders pay through sandwiching and worse fills. This is the visible form, and the reason private RPC endpoints exist.
• Borrowers pay the liquidation bonus, which is set by governance and captured by whoever wins the ordering race.

Framed in the language of transaction costs, extraction is a friction that neither counterparty agreed to and neither can price in advance. Framed as a principal-agent problem, the party choosing the order is not the party bearing the loss, and the loss is invisible in the accounts of the party bearing it.

Why this is a governance problem, not just a trading problem

Three distinct things break for a DAO, and they break in different places.

A vote is an advance announcement of a trade. On-chain governance is a public, timestamped statement of intent with a built-in delay: the proposal names the asset, the size and often the venue, the voting period runs for days, and the timelock adds more. By the time a treasury diversification executes, every searcher on the network has had a week's notice of exactly what is coming. The transparency that makes on-chain governance auditable is the same property that makes its execution predictable, and predictable order flow is the single most valuable input a searcher can have. This is one of the strongest practical arguments for ballot privacy and delayed disclosure – not to hide how members voted, but to stop the treasury telegraphing its own execution.

Extracted value never reaches the fee switch. A DAO's fee switch can only route revenue the protocol actually collected. Value that leaves through arbitrage against the protocol's own pools was never collected in the first place, so it does not appear in any revenue figure a token valuation is built from. A protocol can look structurally unprofitable while generating substantial value, simply because that value is captured one layer below its accounting. Aerodrome makes this explicit in its own design language, describing value normally lost to arbitrage bots as something the protocol should keep.

MEV is now a staking-yield line item, and someone has to decide the split. On Solana, Jito routes MEV rewards to validators and their stakers rather than leaving them with the searcher, which makes the split a governance parameter of a DAO rather than a market outcome. Any DAO holding a liquid staking token therefore holds an exposure to another DAO's MEV policy alongside the slashing exposure it already carries. Osmosis takes the other route and handles ordering at the protocol layer instead of an out-of-protocol auction.

Designs that suppress or redistribute it

No production system eliminates MEV. The designs that work either make ordering worthless, hide the information ordering depends on, or return the proceeds to whoever generated them.

• Batch auctions with uniform clearing prices. CoW Protocol collects signed intents off-chain and auctions them to bonded solvers. Because every order for the same directed pair in a batch settles at the same clearing price, position within the batch stops mattering – the documentation's phrasing is that this “makes transaction order irrelevant”. The auction is combinatorial: solvers bid on individual orders and on groups, and bids that deliver an order less than an available standalone bid are filtered out. This is the strongest structural answer available today, and it is a market-design answer rather than a cryptographic one.
• Private order flow with a rebate. MEV Blocker is an RPC endpoint that withholds transactions from the public mempool, shows them only to searchers who can backrun but not frontrun or sandwich, and pays the user a share of the backrun profit. Its own dashboard reports over $219bn of volume across more than 62m transactions and about 5,500 ETH rebated, at a median rebate of roughly $26 (1 August 2026). MEV-Share runs the same idea from the Flashbots side, selectively revealing transaction details according to the sender's privacy preference and accepting backruns only.
• Enshrining the auction. EIP-7732 moves the builder market into consensus, replacing trusted relays with staked, slashable builders.
• Protocol-level ordering rules. Encrypted or threshold-decrypted mempools, first-come-first-served sequencing, and app-chain-level ordering policy all attack the same input – the searcher's advance sight of your transaction.

Each buys protection with a different cost: batch auctions add latency, private RPCs concentrate order flow in whoever operates the endpoint, and enshrined designs are consensus changes that take years.

What a DAO can actually do about it

Most of the useful moves are operational rather than constitutional, and none of them require a protocol change:

• Never execute a treasury trade from a public mempool. Route it through a solver auction or a private endpoint. For a treasury swap the difference is measured in basis points on a seven-figure order, which is not a rounding error.
• Separate the decision from the execution. Vote on a mandate – the asset, the ceiling, the window – and let the executor choose timing and venue inside it. A proposal that specifies the exact size, venue and block is a proposal that has priced itself in advance.
• Measure LVR on protocol-owned liquidity. Fee income on an LP position is a gross number. The net number is fees minus LVR, and a treasury holding an LP position without measuring the second term does not know whether the position makes money.
• Treat the liquidation bonus as a real parameter. It is set by governance, paid by users, and captured by bots; it deserves the same scrutiny as any other fee.
• Read the MEV policy of any staking token the treasury holds. Who receives the tips, and who can vote to change that, is a governance dependency the treasury has taken on.

How Caper approaches this

Caper does not solve MEV, and no chain-level design here would. What it does is remove one specific surface: the moment a member converts a governance claim into value. In a caper, a member's exit pays out a share of the treasury computed from their own vote weight rather than from an order sold into a pool, and the token leg prices off a deterministic bonding curve rather than a quote that depends on pool depth at the instant of execution. There is no slippage tolerance to widen and no pending order whose size and direction a searcher can read and trade against.

The honest caveat is that ordering still matters wherever a price moves with cumulative volume – being earlier on a rising curve is worth something, and that is true of any curve-priced asset. The claim is narrower and worth stating precisely: the payout a member receives on exit is fixed by a formula over their own holdings and votes, so it is not a trade anyone else can position around.

References and further reading

• Flash Boys 2.0: Frontrunning, Transaction Reordering, and Consensus Instability in Decentralized Exchanges – Daian et al., arXiv:1904.05234, 10 April 2019, the paper that named MEV and formalised priority gas auctions
• Automated Market Making and Loss-Versus-Rebalancing – Milionis, Moallemi, Roughgarden and Zhang, arXiv:2208.06046 (v1 August 2022, v5 May 2024), the closed-form treatment of LP adverse selection
• Flashbots — MEV-Boost and MEV-Share, the reference implementations of out-of-protocol PBS and of backrun refunds
• relayscan.io – live MEV-Boost relay and builder market share, the source of the concentration figures above
• EIP-7732: Enshrined Proposer-Builder Separation – created 28 June 2024, status Review
• CoW Protocol — MEV protection, Fair Combinatorial Batch Auction and Solvers
• MEV Blocker – endpoint behaviour and the published rebate metrics