MEV: The Invisible Economy

:::section{mins=2} What Is MEV :::

:::lead MEV (maximal extractable value) is the extra profit available to whoever decides which transactions go into a block and in what order. Bots collect it, and the bill usually lands on ordinary users and liquidity providers. :::

:::p You swap USDC for ETH. Your wallet quotes 9.87 ETH, and 9.77 arrives. The transaction succeeded and the interface made no mistake. Some of the gap could be ordinary market movement, but there is another possible reason. Between your click and the block, a bot may have spotted your order, bought ETH just before you and sold right after. In that case the bot took most of the difference, and the rest stayed in the pool. :::

:::p{html} That profit is what MEV means. It exists because transactions don't land in a block instantly or in order of arrival. First they wait in the mempool (3.3 §2), a shared list anyone can watch, and then whoever builds the block decides which ones to include and in what sequence. Whoever can influence the order can profit from it. :::

:::p Not all MEV does harm. A bot that wraps itself around your trade took money from you. A bot that brings a pool's price back in line with the exchange price after a big trade made the market more accurate, although the pool's liquidity providers paid for it. Next we'll look at the kinds of MEV, who earns it, and how to avoid becoming its source. :::

:::callout Picture a lock on a busy canal. Barges arrive one after another, and the lock keeper decides who goes through first. Everyone waits in plain sight, and one look at a captain's deck tells you what he's carrying. You're bringing a hold full of strawberries to the morning market in the town past the lock, where the price depends on how much fruit has already arrived. A rival trader notices your cargo, slips the keeper some cash to jump the queue and sells his strawberries first at the morning price. By the time you reach the market, buyers have already stocked up and you have to cut your price. The keeper stole nothing, he only sold a place in line, and the trader profited from knowing what someone else was carrying. On a blockchain the deck is the mempool, the keeper is whoever builds the block, and the cash for jumping the queue is a tip to the validator. :::

:::keypoints{label="Key Points"}

• MEV comes from the right to set transaction order, so it exists on any network where that order is worth money, even where other people's orders aren't visible in advance. html: An MEV victim usually doesn't notice the loss, because the transaction succeeds and the attack only shows up in the neighboring transactions on a block explorer (3.4 §3).
• Some MEV helps the market. Arbitrage, for example, keeps prices aligned across venues, but someone pays for that too, most often liquidity providers. html: Before a large swap on Ethereum, think beyond the rate and the fee to where the transaction is sent, because broadcasting it to the public mempool invites bots (protection options in §6). :::

:::details{badge=Advanced title="How MEV was extracted before Flashbots"} Until 2021, bots fought for block space in the open. When two bots spotted the same opportunity in the mempool, each sent a transaction with a slightly higher gas price, the rival answered with a higher one still, and this went back and forth many times within a few seconds. These races became known as priority gas auctions. Only one bot won, but the losers' transactions still landed in the block, reverted and took up space, so everyone else paid more for gas while it lasted.

The best-known story from that era was told in 2020 by Dan Robinson and Georgios Konstantopoulos of Paradigm in the essay “Ethereum is a Dark Forest”. Someone had accidentally sent about $12,000 worth of tokens straight to a Uniswap pool contract, and whoever called the right function first could take them. The authors tried to rescue the tokens, but a bot spotted their transaction, copied it and took the money for itself.

Bots like this check every transaction in the mempool, and if a copy sent from their own address would turn a profit, they send that copy with a higher fee. That's why Ethereum was compared to a dark forest, where any visible prey instantly becomes a target. The rule that follows still holds today: a transaction that would pay off for any sender must not go through the public mempool. :::

:::details{badge=Advanced title="Why MEV can't simply be banned"} Most MEV follows from how a blockchain works. The price of ETH on exchanges changes every fraction of a second, while the price in a Uniswap pool changes only when someone trades in it. It can move many times within one block, but between blocks, which come every 12 seconds, the pool doesn't react to the outside market on its own. So the pool price keeps lagging behind the market, someone has to correct it, and whoever does so first keeps the difference.

Closing the gap isn't always worth it, because the arbitrageur pays the pool a fee. In a pool with a 0.3% fee and ETH at $3,000 on the market, a trade only becomes profitable once the pool price drops below roughly $2,991 or rises above $3,009. Inside that band the bots sit still, and every time the price breaks out of it, after someone's large trade or a market move, they have work to do.

:::p{html} Banning these bots would leave pools with wrong prices, and without liquidators, lending protocols would pile up undercollateralized debt. So the debate isn't about how to get rid of MEV but about who should get it (§7). ::: :::

:::details{badge=Advanced title="Who pays for “harmless” MEV"} MEV is often split into harmful and harmless. Harmful MEV worsens a specific trade, as a sandwich or front-running does. Harmless MEV doesn't interfere with a trade that has already been sent, and arbitrage after that trade and liquidations fall into this group.

:::p{html} But even harmless MEV has someone paying for it. When an arbitrageur realigns a pool's price, the pool trades with them at a stale price, and liquidity providers cover the difference. In 5.5 §5 this loss is called LVR (loss-versus-rebalancing). So “harmless” only means that you don't pay as a trader. You pay as a liquidity provider, if you've put money into the pool. ::: :::

:::details{badge=Advanced title="How much money flows through MEV"} Nobody knows exactly, because part of MEV is visible on-chain and part is hidden in trades on centralized exchanges. Researchers also measure different things, some bot revenue and others validator income, so the figures below can't be added up or compared directly.

:::p{html}

:::further{label="Further Reading"}

• Flash Boys 2.0: Frontrunning, Transaction Reordering, and Consensus Instability (2019) | arxiv.org
• Ethereum is a Dark Forest (Robinson, Konstantopoulos, 2020) | paradigm.xyz
• Maximal Extractable Value: Implications for crypto markets (ESMA, 2025) | esma.europa.eu :::

:::quiz{id=q-5-6-s1 label="Check yourself" correct=2 explanation="MEV comes from the right to include, exclude and order transactions. Being able to see other people's orders in the mempool adds front-running and sandwiches on top, but arbitrage and liquidations work without it, using state that is already public on-chain. Code bugs lead to hacks, not MEV, fees are only part of the price of block space, and validators get no cut of swaps, since swap fees go to the pool's liquidity providers."} What makes MEV extraction possible on Ethereum?

• Bugs in DEX contracts and developers who are slow to fix them
• High network fees that validators routinely share with bots
• Deciding which transactions go into a block and in what order
• Validators taking a cut of every swap that passes through a DEX :::

:::section{mins=3} Types of MEV :::

:::lead Bots profit from other people's trades by stepping in front of them, around them or right behind them. For an ordinary user the most dangerous one is the sandwich, which can worsen your rate by almost your entire slippage tolerance. :::

:::p The best-known kind of MEV is the sandwich. A bot sees in the mempool that you are buying a token and buys it right before you, which pushes the pool price up, so your purchase goes through at a worse price. Right after you, it sells what it bought at the pushed-up price. Your trade ends up squeezed between two of its trades, like the filling in a sandwich. :::

:::p{html} :::

:::p{html} How much can a sandwich take? Almost as much as your slippage tolerance allows (5.5 §2). When the attack pays off, the bot sizes its purchase so that your trade fills at nearly the worst rate you agreed to, so any tolerance you added "just in case" turns into your loss. Most of it goes to the bot, and the rest stays in the pool. A close relative of the sandwich is front-running, where a bot copies someone else's profitable transaction and pays to have its copy go first. :::

:::p Three other kinds of MEV are more common but don't make your trade worse directly: :::

:::p{html}

• Backrunning. A bot lands right after a large trade and brings the pool price back in line with the market. Your trade has already executed, and the pool's liquidity providers pay for the difference.
• Arbitrage between exchanges. Prices on centralized exchanges change all the time, while the pool price moves only when someone trades in that pool. Between blocks it lags behind the exchange price, and bots profit from the lag. By volume this is one of the largest kinds of MEV (5.4 §7).
• Liquidations. When a borrower's collateral in a lending protocol falls below the threshold, the first bot to repay part of the debt gets the collateral at a discount. The borrower pays for it.

:::callout A Saturday flea market, the row of vinyl stalls. You loudly ask a seller about the first pressing of an album and say you'll take three copies if they come in under 50 euros each. A reseller standing next to you hears this and quickly grabs the two cheapest copies at 30, so you're left with the next ones at 45. A minute later he offers his two to a new buyer, also at 45, since nothing cheaper is left on the table, and that is the whole mechanics of a sandwich. Another trader plays it differently. He waits until you've paid, then brings the same records over from the neighboring market where they sell for 35, and prices between the rows even out, just like arbitrage after a trade. And by the exit stands a third one, waiting for someone to fail to redeem a pawned item at the pawnshop so he can be the first to grab it at a discount. :::

:::keypoints{label="Key Points"}

• A sandwich only pays the bot on trades that noticeably move the pool price, so small swaps in deep pools rarely get attacked. html: A backrun doesn't hurt your trade, but your trade is what creates its profit, and protection RPCs can sell that right and return part of the proceeds to you (§6).
• For the borrower, a liquidation costs the liquidator's bonus, usually a few percent of the repaid debt, and bots strike in the very first block after the threshold is crossed, so counting on topping up your collateral by hand in time is risky.
• Set your slippage tolerance to the price movement you actually expect, not 3-5% just in case, because when the attack pays, the bot will push your fill almost down to the minimum you set. :::

:::details{badge=Advanced title="The sandwich from the diagram, in numbers"} The diagram in the main text uses a simple example. The pool holds 1000 ETH and 3 million USDC, so ETH is priced at $3000, and the pool fee is 0.3%. You swap 30,000 USDC and, with no bot around, would get 9.87 ETH. That's slightly worse than $3000 per ETH because the pool is shallow and your own trade moves the price. This is price impact. It is already built into the quote, and the bot worsens your fill on top of it.

You set a 1% tolerance, which means you accept no less than 9.77 ETH. The bot sees this and buys ETH with 15,211 USDC right before you, exactly enough to leave you with 9.77. Right after your trade it sells the 5.03 ETH it bought and gets 15,422 USDC, making about $210. You receive 0.1 ETH less, about $296, and the rest of the difference comes from the fees the bot paid to the pool and the change in the value of the pool's reserves.

Run the same numbers with a different tolerance and a pattern shows up. At 0.5% you lose about $148, and at 3% about $888. The loss is almost exactly the tolerance times the trade size, so your tolerance is the ceiling on what a sandwich can take. :::

:::details{badge=Advanced title="When a sandwich doesn't pay off"} The bot pays the pool fee twice, once on the buy and once on the sell, so the attack only makes sense if your trade moves the price by more than those two fees. In our pool with a 0.3% fee, attacking a 5000 USDC swap is unprofitable at any tolerance, and a 10,000 USDC swap would earn the bot only about $9.

Low-fee pools turn this around. In a pool with a 0.05% fee, the same 5000 USDC swap already gives the bot about $35, and in stablecoin pools with a 0.01% fee the bot loses almost nothing to fees. According to EigenPhi, about 12% of sandwiches on Ethereum hit stablecoin swaps, so a low pool fee saves you money but also makes you a convenient target. :::

:::details{badge=Advanced title="Who pays for the arbitrage after your trade"} Back to the 30,000 USDC swap, this time without a bot. Afterwards ETH costs about $3060 in the pool but still $3000 on exchanges. An arbitrageur sells 8.41 ETH into the pool for 25,447 USDC, while the same ETH is worth $25,233 on an exchange, so the profit is about $214.

The pool price then settles at about $3009 rather than $3000, because the fee makes further trading unprofitable. Your trade has already executed, and the $214 is paid not by you but by the liquidity providers, since the pool bought ETH from the arbitrageur above the market price. :::

:::details{badge=Advanced title="A liquidation in numbers"} Say a borrower deposits 5 ETH into a lending protocol when ETH is at $3000 and borrows 11,000 USDC. The protocol allows the position to be liquidated once the debt exceeds 82.5% of the collateral value, and in this example that happens if ETH falls to about $2667.

:::p{html} If ETH drops to $2600, a liquidator bot repays half the debt, 5500 USDC, and receives $5775 worth of ETH for it, which is a 5% bonus. The bot earns $275 minus gas, and the borrower loses the same $275, giving up $5775 of collateral to clear $5500 of debt. This bonus is what bots race for in every block after a sharp price drop (6.3 §5). ::: :::

:::details{badge=Advanced title="JIT liquidity"} One kind of MEV hits liquidity providers rather than traders. JIT (just-in-time) liquidity works in concentrated-liquidity pools such as Uniswap V3. The bot spots a large swap in the mempool, adds a lot of liquidity in a narrow range around the current price right before it, and pulls it out right after the swap.

Swap fees are split among providers in proportion to their liquidity in the active range, so for one block the bot becomes the dominant provider and takes most of the fee from that trade. The trader loses nothing, but regular providers who keep their money in the pool for weeks earn less from the big trade than they counted on. :::

:::details{badge=Advanced title="How the most famous sandwich bot was beaten at its own game"} Ethereum's best-known sandwich bot is jaredfromsubway.eth, active since early 2023. According to EigenPhi, from November 2024 to October 2025 it accounted for about 70% of all sandwiches on the network. Making money is getting harder for it, since the bot hands most of what it takes to builders for the right spot in the block, and in April 2025 it even ran at a loss.

On June 20, 2026, the bot walked into a trap. For several weeks an attacker had been creating fake tokens resembling WETH, USDC and USDT, along with pools that looked profitable to the bot. Small trades went through honestly, the bot learned to trust them and granted those contracts approvals to spend its tokens, and at the right moment the attacker used the approvals to drain about $7.5 million from the bot. An automated strategy that trusts any token with a familiar-looking name is as vulnerable as a careless person. :::

:::further{label="Further Reading"}

• Maximal extractable value (MEV) on ethereum.org | ethereum.org
• EigenPhi data on the decline of sandwich attacks on Ethereum (2024-2025) | cointelegraph.com
• The predator becomes the prey (how jaredfromsubway was drained) | blockaid.io :::

:::quiz{id=q-5-6-s2 label="Check yourself" correct=1 explanation="Your tolerance sets the worst rate you'll accept, and the bot can read it right in the transaction. It sizes its purchase so your trade fills at almost exactly that worst rate, so almost the whole gap between the quote and your limit can be lost. Most of the loss goes to the bot and the rest stays in the pool. In the worked example you lose about $296 while the bot makes about $210. A narrow tolerance leaves the bot less room, and if it's too narrow, the trade simply reverts when the price moves on its own."} How is the slippage tolerance you set on a swap related to a sandwich?

• The bot doesn't care, since tolerance only affects the pool fee
• The wider the tolerance, the more the bot can take from your trade
• A wide tolerance scares the bot off because it can't tell your limit
• A narrow tolerance suits the bot better because the pool price stays put :::

:::section{mins=2} Searchers, Builders and Proposers :::

:::lead Several more players stand between the bot that spots a profit and the finished block. Every step runs an auction, so most of the money ends up with the validator who gets to propose the block. :::

:::p Who actually makes money on a sandwich? It looks like the bot, but on Ethereum since 2022 most of the money moves further down a chain with four roles: :::

:::p{html}

• Searcher. A bot that hunts for profit and packs it into a bundle, a set of transactions that execute only together and in a fixed order. It offers a payment to get the bundle included.
• Builder. Assembles the most profitable block it can from bundles and ordinary transactions, then bids for the right to have the validator publish it.
• Relay. A middleman that shows the validator only the bid and reveals the block after it is signed, so the validator cannot steal other people's strategies.
• Proposer. The validator assigned to this slot (2.2 §5). It picks the highest bid and signs the block.

:::p{html} :::

:::p Why does the proposer get the most? Because every step is an auction. If five bots see the same arbitrage, they outbid each other until almost nothing is left for themselves, and builders do the same when they fight for the slot. The money settles with whoever has no competitor in that slot. :::

:::p The setup has a downside. The winning builder tends to be the one with the most exclusive order flow, so block building is concentrated in a few companies. In the 24 hours to October 21, 2026, relayscan data shows that the builder Titan produced about half of all Ethereum blocks. :::

:::callout A summer street-food festival in a city park. Cooks from all over town turn up with ideas for dishes and are ready to give up part of their takings for a good spot. The festival organizer turns these applications into the lineup that will bring in the most money and decides who stands by the entrance and who ends up in the far corner. But the park does not belong to the organizer. Every Saturday the park administration rents the grounds to whichever organizer offers the most. So that the administration cannot peek at someone else's lineup and run the festival itself, applications go through an agent who reveals only the amount. In this chain the cooks are the searchers, the organizer is the builder, the agent is the relay, and the park administration is the proposer. And as so often happens in life, the owner of the venue walks away with the most. :::

:::keypoints{label="Key Points"}

• Competition between bots does not make a sandwich any cheaper for the victim. It only changes who walks away with what was taken, so the thing to defend against is the attack itself.
• MEV income is part of staking yield, so some of the return on staked ETH depends on how active the bots are and drops in quiet months.
• The biggest edge in this market is order flow that nobody else can see, which is why wallets and aggregators have become sought-after partners for builders.
• When a wallet or service offers “private sending”, you are choosing who sees your order before it reaches a block, so it is worth checking who runs that service. :::

:::details{badge=Advanced title="How the players pay each other"} :::p{html} :::

:::p{html} Inside the chain, money moves step by step. The bot pays the builder for its bundle's spot, either through the gas fee or with a direct transfer inside the bundle, and the transfer has a handy property: it only goes through if the whole bundle succeeds. The builder collects all of the block's fees and, in the last transaction, sends the validator the promised bid. The base fee is burned along the way and goes to no one (2.2 §3). ::: :::

:::details{badge=Advanced title="Who keeps how much"} :::p{html} :::

Take a hypothetical example. A builder has assembled a block that earns it 0.20 ETH, and a rival can bid 0.19 ETH for its own block. Our builder only needs to bid slightly more, say 0.195 ETH, and keeps just 0.005 ETH for itself.

Real data looks much the same. A study of arbitrage between exchanges from August 2023 to March 2025 found that bots handed builders about 60% of their revenue, and bots owned by the same companies as the builders handed over up to 90%. Builders themselves, according to relayscan for the 24 hours to October 22, 2026, kept about 0.012 ETH per block on average, and the rest went to validators. The fiercer the competition at a link, the less money stays there. :::

:::details{badge=Advanced title="Why big companies win"} :::p{html} :::

A builder wins the auction when it can put together a more valuable block than its rivals. For that it needs transactions nobody else has, such as order flow from a popular wallet or from its own trading bot. That is why big companies keep the bot and the builder under one roof, and the bot hands its profit not to a stranger but to itself.

This is how profit gathers in a few hands. In the 24 hours to October 22, 2026, Titan built about 54% of Ethereum blocks but took about 70% of all builder profit. That is a risk for the network, because a handful of companies end up deciding which transactions make it into most blocks. :::

:::details{badge=Advanced title="Why the relay has to be trusted"} :::p{html} :::

The relay exists so that the validator and the builder cannot cheat each other. The validator does not see the block before signing, so it cannot steal the bots' strategies, and the builder cannot skip payment because the relay has checked the bid in advance. But the relay itself is not part of the protocol, works for free and runs on reputation.

On April 3, 2023, someone exploited that trust. A validator signed a block header, used a bug in the relay's code to see the contents early and replaced the sandwich bots' transactions with its own, and the bots lost about $25 million in a single slot. According to prosecutors, the validator was run by brothers Anton and James Peraire-Bueno, and in May 2024 the US Department of Justice charged them with fraud. The first trial, in November 2025, ended without a verdict because the jury could not agree.

:::p{html} This episode strengthened the case for ePBS, where the block-for-payment exchange follows protocol rules with no middleman (§4). ::: :::

:::further{label="Further Reading"}

• Live relay and builder market shares on Relayscan | relayscan.io
• How MEV is split between searchers, builders and validators (EigenPhi) | eigenphi.substack.com
• Measuring CEX-DEX Extracted Value and Searcher Profitability (2025) | arxiv.org :::

:::quiz{id=q-5-6-s3 label="Check yourself" correct=3 explanation="Every link in the chain runs an auction. Searchers hand almost all their profit to builders, and builders hand almost the entire block value to the proposer, who has no competitor in its own slot. The protocol sets no fixed share. With MEV-Boost the proposer usually does not build the block itself, and relays distribute nothing, since the builder pays the bid to the proposer directly inside the block and only that slot's proposer receives it."} Why does most MEV on Ethereum end up with the proposer?

• The Ethereum protocol assigns it a fixed share of MEV
• The proposer runs its own bots and builds the block with no middlemen
• Relays split MEV among validators in proportion to their stake
• Searchers and builders outbid each other down to a near-zero margin :::

:::section{mins=2} Flashbots, MEV-Boost and PBS :::

:::lead Until 2021, bots fought for block space in the open, bidding up fees against each other and clogging the network. Today that space is sold in a sealed auction, the validator simply takes the highest bid, and that is how the vast majority of Ethereum blocks get built. :::

:::p In 2020, MEV on Ethereum looked like a crowd pushing at a checkout. Bots spotted the same opportunity and raced each other to raise their gas price. One of them won, and the others' transactions reverted but still took up space in blocks, so ordinary users overpaid for every action while this went on. :::

:::p Flashbots, a research organization founded in 2020, moved the fight into a sealed auction. A bot sends its bundle straight to a miner (after the move to Proof of Stake, to a block builder) and states how much it is willing to pay, while losing bundles simply never make it into a block and cost nothing. The network calmed down, even though MEV itself did not go away. :::

:::p After the move to Proof of Stake, the roles were split even further, and the idea got a name, PBS (proposer-builder separation). The validator runs a program called MEV-Boost, receives bids from builders and picks the highest one, while a specialist builds the block. This way a home validator earns as much from MEV as a data center, and 85-95% of blocks go through MEV-Boost. Ethereum's next step, the ePBS upgrade, will build this scheme into the protocol itself, so nobody has to trust intermediaries. :::

:::callout Open mic night at a comedy club. At first the running order was settled by shouting. Comics crowded the stage, talked over each other and promised the host more and more money, and the audience spent half an hour listening to the squabble instead of jokes. Then the club introduced sealed notes, where everyone writes down how much they will pay for a slot, and the host quietly picks the best ones. The noise left the room and the evening became predictable. When the club got an owner who knows nothing about comedy, he stopped putting the lineup together himself and now hands each week's show to whichever booker offers the most for it. The notes play the role of Flashbots bundles, the booker works like a builder, and the club owner like the proposer in PBS. :::

:::keypoints{label="Key Points"}

• Sealed auctions kept losing bids out of blocks and took the load off the network, but they did not shrink MEV itself, so sandwiches and arbitrage still cost users money.
• PBS evened out income between small and large validators, but block building concentrated among a few builders, and they are now the ones who decide what goes into most blocks.
• Censorship on Ethereum usually means a delay rather than a ban, because a filtered transaction waits for a block from a participant who does not filter.
• If you run your own validator, choose your MEV-Boost relays deliberately, because that list decides whether your blocks filter transactions and whom you are trusting. :::

:::details{badge=Advanced title="How the auction for one slot works"} A new Ethereum block appears every 12 seconds, and the auction for it runs the whole time. Builders keep rebuilding their blocks with fresh transactions and send them to relays together with a bid. When the moment comes, the validator asks the relays for the highest bid, signs that block without seeing its contents, and the relay publishes the block to the network.

:::p{html} :::

:::p{html} Why does the validator sign blind? If it could see the contents, it could take other people's bundles for itself and pay the builder nothing. The signature commits it to publishing exactly this block, and a validator that signs a different one in the same slot gets slashed (1.3 §3). ::: :::

:::details{badge=Advanced title="Where this setup is weak"} :::p{html} The first weak spot is the relays themselves. There are only a handful of them, they are not part of the Ethereum protocol, and they run for free on their operators' reputation. If a relay makes a mistake, a validator can see someone else's block too early, and in April 2023 that is exactly how sandwich bots lost about $25 million (§3). :::

The second weak spot is censorship. In 2022 the US added the Tornado Cash mixer to its sanctions list, and some relays started dropping transactions linked to it. By the end of that year, 70-80% of blocks went through such relays. The transactions did not vanish, though. They waited for a block from a validator or builder without filters, so censorship meant a delay rather than a ban. The sanctions were lifted in March 2025, but protection against censorship still rests only on neutral participants outnumbering the rest. :::

:::details{badge=Advanced title="What ePBS and FOCIL change"} Both changes move rules out of informal arrangements and into the protocol itself. ePBS (EIP-7732) makes the relay's job part of Ethereum. The validator picks a builder's bid under network rules, and the builder's collateral guarantees the payment, so there is no longer a middleman you have to trust. MEV doesn't go away, though, and block building stays with a handful of large builders.

FOCIL (EIP-7805) deals with censorship. In every slot a small group of validators publishes a list of transactions they see in the mempool, and the block has to include them if it has room. The builder still decides the order, but it can no longer throw a transaction out.

ePBS is part of the Glamsterdam upgrade, which has been running on a testnet since October 6, 2026, and no mainnet date has been announced yet. FOCIL is planned for the following upgrade, Hegotá, expected in 2027. In short, block space is sold today through intermediaries that run on trust, and Ethereum is gradually moving this market onto protocol rules. :::

:::further{label="Further Reading"}

• Glamsterdam Testnet Announcement (ePBS, EIP-7732) | blog.ethereum.org
• EIP-7805: Fork-choice enforced Inclusion Lists (FOCIL) | eips.ethereum.org
• MEV-Boost overview from Flashbots | docs.flashbots.net :::

:::quiz{id=q-5-6-s4 label="Check yourself" correct=2 explanation="MEV-Boost connects the validator to the builders' auction. Builders put together the most profitable block and pay for the right to publish it, and the validator just picks the highest bid. It does not hide any transactions or run bots, and swap fees go to liquidity providers, not to validators."} Why does a validator run MEV-Boost?

• To hide its own transactions from bots watching the mempool
• To run bots itself and profit from sandwich attacks
• To earn more per block by letting specialists build it
• To get a cut of the fees from every DEX swap :::

:::section{mins=2} MEV Across Networks :::

:::lead MEV shows up wherever transaction order is worth money, but it looks different on every network. Ethereum sells block space through an auction, Solana through tips to the validator, and on L2s the sequencer sets the order, so the main problem there is not sandwiches but bot spam. :::

:::p{html} Solana has no shared public mempool. Transactions go straight to the validator that is currently producing blocks (2.3 §3). For a long time people assumed this protected users from sandwiches. But a validator can sell access to the transactions it receives, and that is how private mempools appeared, where bots saw other people's trades before they executed. According to a Helius report, one such program made about $13 million from sandwiches over 30 days around the turn of 2024 and 2025. Since June 2024 the Solana Foundation has been withdrawing its delegated stake from validators caught doing this. :::

:::p{html} On L2s such as Arbitrum and Base, order is set by the sequencer, a single server run by the network operator (2.5 §8). It does not reveal transactions in advance, so sandwiches are rare. The catch is that bots have no way to buy a spot right next to a profitable trade, so they fire off thousands of probe transactions hoping to catch an arbitrage. Flashbots estimated that in February 2025 this spam took up more than half of the gas in Base blocks. :::

:::p{html}

:::callout Three nightclubs in the same city. At the first, you get in through promoters who sell spots on the guest list and bid against each other for the best hours, and the owner collects the highest price for the night. The second has no list, but the bouncer at the door changes every half hour, and you can strike a deal with each one directly by slipping him a bill. At the third, the owner stands at the door himself and lets people in strictly first come, first served, so a pushy neighbor plants a dozen friends by the entrance to make sure one of them is first when a spot opens, and the line stretches round the block. The first club is Ethereum with its builder auction, the second is Solana with tips to the slot leader, and the third is an L2 with a single sequencer and bot spam. :::

:::keypoints{label="Key Points"}

• Having no public mempool does not protect against MEV by itself. It only hands access to transactions to whoever sits closest to the validator.
• On L2s the cost of MEV is paid not by individual sandwich victims but by every user of the network, because bot probe transactions fill up block space.
• The more networks there are, the more their prices drift apart, so a single large trade on Ethereum creates work for bots on a dozen networks at once.
• For sizeable swaps on Solana, pick an interface with sandwich protection and a tight slippage tolerance, because your trade can be seen before it executes. :::

:::details{badge=Advanced title="How Jito Works"} On Solana a new slot begins roughly every 400 milliseconds, and one validator produces blocks for four slots in a row. Most validators run the Jito client, which holds a small auction before every block. Bots send in bundles with tips in SOL attached, and Jito picks the winning bundles that do not conflict with each other and passes them to the validator.

Jito used to run its own mempool too, where bots saw ordinary user transactions 200 milliseconds before execution and built sandwiches around them. Jito shut it down in March 2024, even though that cost it a noticeable share of revenue. According to Blockworks Advisory, by the end of Q2 2026 Jito clients ran on about 54% of Solana stake, and roughly a third of all stake had already moved to BAM, the new version, where transaction order follows rules announced in advance. :::

:::details{badge=Advanced title="Sandwiches on Solana"} Closing the Jito mempool did not stop sandwiches. They moved into the private mempools of individual validators. Helius data shows that a program linked to the operator DeezNode carried out 1.55 million sandwich transactions over 30 days from December 2024 to January 2025 and earned about 65,880 SOL, roughly $13.4 million, or about $8.7 per attack on average.

Memecoins became especially easy prey. Their pools are shallow, prices jump around, and interfaces often suggest a wide slippage tolerance just so the trade goes through, while a sandwich can worsen your rate by almost that entire tolerance. In June 2024 the Solana Foundation began removing validators caught doing this from its delegation program, but it cannot ban private mempools. All it can do is withdraw its stake. :::

:::details{badge=Advanced title="Why L2s Get Spam Instead of Sandwiches"} An L2 sequencer orders transactions either by arrival time or by the size of the priority fee, as networks built on the OP Stack do. It shows them to no one in advance, so a bot cannot pick a spot right after someone else's trade. What is left is to send lots of probe transactions, each of which checks at execution time whether a profitable arbitrage exists and quietly exits if it does not.

Flashbots estimated that in February 2025 these transactions used about 56% of the gas in Base blocks but paid only about 14% of the fees, and two firms produced more than 80% of this spam. It does not hurt users directly, but the network spends block space on junk. :::

:::details{badge=Advanced title="How Arbitrum Sells Priority"} Arbitrum decided to sell places in the queue openly. Since April 2025 it has run Timeboost, an auction for an "express lane" held once a minute. The winner's transactions go to the sequencer immediately, everyone else's are delayed by 200 milliseconds, and the auction proceeds go to the network's DAO.

The outcome was not what the designers intended. According to a 2025 study, by mid-September the DAO had earned about 1,090 ETH, roughly $4.9 million, but two firms, Selini Capital and Wintermute, won more than 90% of the auctions, and spam did not go down. Selling priority turns MEV into network revenue, but the same few professionals win it almost every time. :::

:::details{badge=Advanced title="Cross-Chain MEV"} With many networks, prices between them keep drifting apart. A large ETH sale on Ethereum knocks down the price in mainnet pools, while pools on Arbitrum and Base still show the old price for a while, and someone buys the cheaper ETH on Ethereum and sells it on the L2, bringing the prices back in line.

:::p{html} This cannot be done in a single transaction, because different networks share no common block. So this kind of arbitrage is done by players who keep funds on every network in advance or use fast bridges (3.5 §3). A 2025 study estimated that from September 2023 to August 2024 cross-chain arbitrage made about $9.5 million in profit. ::: :::

:::further{label="Further Reading"}

• Solana MEV Report: Trends, Insights, and Challenges (2025) | helius.dev
• MEV and the Limits of Scaling (Flashbots, 2025) | writings.flashbots.net
• The Express Lane to Spam and Centralization (a Timeboost analysis) | arxiv.org :::

:::quiz{id=q-5-6-s5 label="Check yourself" correct=2 explanation="On Solana, transactions go straight to the validator producing the block, and it decides who gets to see them. Some validators sold that access to bots through private mempools. Neither how long transactions stay pending nor the protocol itself has anything to do with it, and wallet hacks are a different kind of attack, not MEV."} Why do sandwiches happen on Solana even though it has no shared public mempool?

• Solana keeps pending transactions public longer than Ethereum does
• Bots on Solana hack wallets and swap out transactions
• A validator can sell bots access to other users' transactions
• Sandwiches are built into the Solana protocol for validators by default :::

:::section{mins=3} How to Protect Yourself from MEV :::

:::lead On Ethereum you can shield yourself from sandwiches almost completely, for free, just by keeping your transaction out of the public mempool. MEV Blocker, one of the protection RPCs, had routed more than $219 billion in trades and returned over 5,500 ETH to users by October 2026, yet no single tool covers every kind of MEV. :::

:::p A sandwich only works if a bot sees your transaction before it lands in a block. So the most effective protection, and a free one, is to send transactions through a private RPC instead of the public mempool. That is a node which passes them straight to builders and limits who can use them and how. Every tool has its limits, though, so it helps to know the whole kit: :::

:::p{html}

• Private RPC. Flashbots Protect or MEV Blocker forward your transaction directly to builders, and bots watching the mempool never see it. A sandwich becomes practically impossible. Backrunning still happens, but the protection RPC sells the right to do it and returns most of the proceeds to you.
• Intents and batch auctions. In CoW Swap, UniswapX or 1inch Fusion you sign an intent rather than a transaction, something like "I give X and want at least Y", and solvers compete to find the best route. In CoW Swap every trade in the same pair and direction within an auction settles at one uniform price, so the order inside the auction does not matter, and the solver's gas is already priced into your rate.
• Slippage tolerance. The tighter it is, the less a sandwich can take. Set it too tight and the transaction reverts on ordinary price moves, and if you sent it through the public mempool you still pay gas for the failed attempt.
• Choice of chain and pool. The deeper the pool relative to your trade, the less your trade moves the price and the less it interests a bot (the rule from §2). On an L2 whose sequencer does not reveal transactions in advance, sandwiches are rare.

:::p Does a private RPC mean you can stop thinking about MEV? No. Instead of every bot in the world, your order is now seen by the RPC operator and the builders it forwards to, so trust moves rather than disappears. A batch auction protects you from the classic sandwich, but with only a few solvers competing, the final rate can be worse than the best on the market, and you lose out with no bot involved at all. And no RPC will save a position you have pushed to its liquidation threshold yourself, because there the bot is not attacking your transaction, it is enforcing the protocol's rules. :::

:::callout A summer night camping by a lake, with more mosquitoes than stars. You can pitch a tent and zip the door shut, and almost nothing gets in, though you are trusting the zipper and the seams. You can string a shared net over the table so the whole group eats under it at once, and then the mosquitoes no longer care who sat down first. You can put on repellent, which will not stop the most persistent ones but makes you a poor target. Or you can pick a windy point for your campsite from the start, where there are fewer mosquitoes anyway. An experienced camper does all of it and still packs bite cream. MEV works the same way: the tent is a private RPC, the net is a batch auction, the repellent is a tight slippage tolerance, and the windy point is a chain and pool where sandwiches do not pay. :::

:::keypoints{label="Key Points"}

• For an ordinary swap, the main benefit of a private RPC is that sandwich risk becomes very low, while backrun refunds only add up on large trades, so choose an RPC for its protection and the operator's reputation, not for promised payouts.
• With intents, the quality of your rate depends on how many solvers compete, so on thin pairs and in quiet hours compare the result with a direct swap and an aggregator instead of assuming the intent is the better deal.
• Private routing only protects you on one chain: an Ethereum RPC does nothing for your transactions on Solana or an L2, and for each chain you need to know separately who can see your order before the block.
• A good everyday habit: a protection RPC for Ethereum or your wallet's built-in private sending, a slippage tolerance that fits the trade, and a look at the minimum amount the interface guarantees right before you sign. :::

:::details{badge=Advanced title="What happens to a transaction in a private RPC"} Your wallet sends signed transactions through an RPC, the address of a node that passes them on to the network. A regular RPC publishes the transaction in the shared mempool, where everyone can see it. A protection RPC does not publish it. It hands it directly to a few builders it trusts, and if a different builder won the block in that slot, the transaction waits for the next one, so confirmation sometimes arrives a few seconds later.

Flashbots Protect also guards against failed transactions: a transaction only goes into a block if it does not revert, so you pay no gas for a failed attempt. To connect such an RPC, replace the address in your wallet's Ethereum network settings, for example with rpc.flashbots.net/fast or rpc.mevblocker.io. Some wallets already have private sending built in. :::

:::details{badge=Advanced title="How backrun refunds work"} MEV Blocker, developed by CoW DAO, shows your transaction to connected bots but only lets them go after it, never before. The bots bid for the right to backrun, 90% of the winning bid comes back to you in the same block, and the validator gets the other 10%. If backrunning your swap is worth 0.01 ETH, you receive about 0.009 ETH on top.

As of October 2026, the project's site reports more than $219 billion in trades and 5,500 ETH returned to users, with a median refund of about $26. The developers themselves still recommend a sensible slippage tolerance, because no protection blocks every sandwich. :::

:::details{badge=Advanced title="How the CoW batch auction works"} In CoW Swap you do not send the swap to a pool yourself. You sign an order, and solvers compete to offer the best result. All orders in the same pair and direction within an auction settle at a single price, so the order of transactions in the block makes no difference and a sandwich has nothing to grab onto.

When two orders are opposite, say one person sells ETH for USDC and another buys, the solver matches them directly, and neither side pays the pool fee or slippage. The downside is that price quality depends on how many solvers take part, and their gas is already in the rate, so small orders are sometimes cheaper as a regular swap on a low-fee chain. :::

:::details{badge=Advanced title="How to set your slippage tolerance"} :::p{html} The example in §2 shows that a sandwich can worsen your rate by almost the entire tolerance, so set it according to how far the price could realistically move while you wait for a block. Stablecoin pairs usually need only hundredths of a percent, liquid pairs like ETH/USDC often do fine with around 0.5%, and a wide tolerance is only needed for volatile tokens with shallow pools, which is exactly where private sending matters most. :::

Tokens with a hidden transfer fee are a separate case. They need a wider tolerance for technical reasons, which is all the more reason to watch carefully where the transaction goes. :::

:::details{badge=Advanced title="Where protection does not help"} :::p{html} A private RPC will not save you from liquidation, because there the bot is not attacking your transaction but enforcing the protocol's rules, and your only defense is a healthy collateral buffer. It will not help if you picked a shallow pool with a bad price yourself, and it will not protect you from phishing signatures, which have nothing to do with MEV (4.1 §2). And it only works on the chain it is set up for, while Solana and L2s have their own rules (§5). ::: :::

:::further{label="Further Reading"}

• Flashbots Protect: how to connect it and what it protects against | docs.flashbots.net
• MEV Blocker RPC modes and refund statistics | mevblocker.io
• How CoW Protocol batch auctions work | docs.cow.fi :::

:::quiz{id=q-5-6-s6 label="Check yourself" correct=3 explanation="A private RPC keeps the transaction out of the public mempool, so sandwich risk drops sharply, although it still depends on the service's rules, its participants and the absence of leaks. Liquidation has nothing to do with how you send a transaction, because there the bot enforces the protocol's rules rather than attacking your trade. Slippage does not go away either, since the pool price still moves with other trades, and now you trust the RPC operator and the builders it forwards your order to."} You sent a swap on Ethereum through a private RPC. What did that protect you from?

• Liquidation of your position if the collateral price crashes
• Slippage, since the rate is locked at the moment you sign
• Having to trust anyone other than the pool contract
• Most sandwich risk, by skipping the public mempool :::

:::section{mins=2} The Future of MEV :::

:::lead MEV can't be abolished, because markets need arbitrage and liquidations, but who gets it is still open to choice. Ethereum is moving the rules of the block auction into the protocol itself, and applications are learning to take MEV away from bots and hand it back to users. :::

:::p Could MEV be eliminated altogether? Not really. Prices on exchanges move all the time, while a pool's price only moves when someone trades in that pool, so somebody will always earn by bringing the two back in line. And as long as collateralized loans exist, somebody will race to liquidate them first. So the debate about the future is about something else: who gets the MEV. The bots, the validators, the protocol, or the user whose trade created it. Four directions are being developed right now: :::

:::p{html}

• Hide the transaction until it is in a block. In an encrypted mempool nobody can read a transaction until its place in the block is fixed, so a bot has nothing to aim at. Gnosis Chain already works this way; for Ethereum it is still research.
• Build blocks in a sealed environment. BuilderNet assembles blocks inside a protected area of the processor that even the server owner can't look into, so nobody sees other people's orders.
• Return MEV to whoever created it. Protection RPCs already refund users part of the backrun revenue (§6), and Chainlink SVR does the same for lending protocols with the money from liquidations.
• Write the rules into the protocol. Ethereum is building the block auction into the protocol (ePBS) and adding forced transaction inclusion (FOCIL, §4) so it depends less on intermediaries.

:::p Meanwhile the way people trade is changing too. More and more swaps go through intents: you describe what you want to receive, and solvers find the route. Instead of bots hunting for your transaction in the mempool, solvers compete for the right to fill your order. :::

:::callout You can't ban the wind on a coastline. It comes from differences in pressure and will keep blowing as long as the sun heats land and sea unevenly. A hundred years ago people fought it with fences and shutters. Today wind turbines line the shore, and the same force spins them and pays for the village's electricity. MEV is like that wind: it comes from price gaps between markets and from the fact that blocks arrive one after another rather than continuously. Encrypted mempools work like shutters and protect a single house from the gusts. Refunds, order flow auctions and protocols capturing their own MEV are the turbines. They don't stop the wind, they decide who gets its energy. :::

:::keypoints{label="Key Points"}

• Protection and redistribution solve different problems. Encryption reduces harmful MEV, while auctions and refunds share out the useful kind, and a mature system needs both at once.
• Every new ordering mechanism adds its own point of trust, whether that is the chip maker behind a TEE, the keyholders of an encrypted mempool or the oracle that intercepts liquidations, so the risks move rather than disappear.
• The debate around FOCIL is about responsibility as much as engineering. The more strictly the protocol forces transactions in, the less choice a validator has, and that changes the picture for regulators.
• When you size up a new DEX or L2, ask who decides the order of transactions and who gets the MEV. An honest answer tells you more about execution than any "MEV-free" marketing promise. :::

:::details{badge=Advanced title="How an encrypted mempool works"} The idea is that nobody can read a transaction until its place in the block is locked in. The user encrypts the transaction, and the key is split among a group of participants. It can only be reconstructed if enough of them combine their shares. First the encrypted transactions go into a block, then the group publishes the key, and only then are the transactions decrypted and executed.

There are weak spots too. If the keyholders collude, they can read transactions early, and if many of them go offline, transactions get stuck. On top of that, encryption hides the contents but not the sender or the size of the transaction. Gnosis Chain launched a mempool like this, based on Shutter, at the end of 2024; for Ethereum it is still research. :::

:::details{badge=Advanced title="Building blocks inside secure hardware"} Modern processors can run a program in an isolated environment (TEE, trusted execution environment) that even the server owner can't look into, and prove to the outside world exactly which program is running there. BuilderNet, launched by Flashbots, Beaverbuild and Nethermind in November 2024, builds blocks inside such an environment, so the operator can't see other people's orders or change the rules. In the 24 hours before October 21, 2026, BuilderNet built about 16% of Ethereum blocks.

The downside is that you have to trust the chip maker, and hardware vulnerabilities turn up in these environments regularly. That is why TEEs are seen as an intermediate step: better than trusting a single company, but weaker than mathematical guarantees. :::

:::details{badge=Advanced title="How an application can capture MEV for itself"} On chains where transactions are strictly ordered by priority fee, such as L2s built on the OP Stack, an application can charge each transaction an extra fee that is a multiple of that priority fee. Dan Robinson and Dave White of Paradigm described the idea in 2024 in the paper "Priority Is All You Need".

Say the extra fee is 99 times the priority fee and an arbitrage is worth $100. Bots keep raising their bids until their costs approach $100, and every dollar of priority fee costs them $100 once the extra fee is included, so they stop at roughly $1 of priority fee. The network then gets $1 and the application about $99. The scheme only works if whoever orders the transactions plays by the rules honestly. :::

:::details{badge=Advanced title="Returning MEV from liquidations"} Normally the liquidation bonus goes to bots and validators. Chainlink SVR (Smart Value Recapture) works differently: the oracle price update that makes a position liquidatable first goes to a private auction. Bots bid for the right to run the liquidation right after it, and the protocol collects the proceeds.

As of May 2026, SVR had recaptured about $18.7 million, of which lending protocols received about $12 million and Chainlink about $6.7 million. Liquidation is no cheaper for the borrower, but the bonus stays inside the protocol. :::

:::details{badge=Advanced title="What to expect in the next few years"} As of late October 2026, the Glamsterdam upgrade with ePBS is running on a testnet, and no mainnet date has been announced. The next upgrade, Hegotá, with FOCIL inclusion lists, is expected in 2027, and both timelines may slip. MEV will most likely become less visible to ordinary users, as more trades go through private channels and intents, and the fight over who controls transaction ordering will move into protocol design. :::

:::further{label="Further Reading"}

• Priority Is All You Need (Robinson, White, 2024) | paradigm.xyz
• BuilderNet documentation | buildernet.org
• Chainlink Smart Value Recapture (SVR) | blog.chain.link :::

:::quiz{id=q-5-6-s7 label="Check yourself" correct=0 explanation="Auctions like MEV Blocker or Chainlink SVR leave arbitrage and liquidations in place but make bots pay for the right to execute them, and that money goes back to the user or the protocol. An encrypted mempool cuts harmful MEV but doesn't redistribute anything by itself: the arbitrage after decryption still goes to bots. Slower blocks would only widen price gaps and increase MEV, and moving to a centralized exchange swaps the blockchain for trust in an intermediary that has its own ways to profit from order flow."} Which approach doesn't destroy MEV but hands it to the protocol or the user?

• Auctioning the right to backrun or liquidate and refunding the proceeds
• Making blocks slower so that bots no longer have time to react to trades
• An encrypted mempool that hides what transactions contain
• Moving swaps to centralized exchanges with an order book :::