Skip to content
DeFi Reference

MEV supply chain

1. TL;DR

Section titled “1. TL;DR”

On Ethereum the right to order a block passes through a public mempool, searchers who assemble bundles, builders who assemble full blocks, relays that escrow them, and the slot’s proposer, who signs the highest-paying header without seeing its contents. On Solana there is no in-protocol mempool: transactions go straight to the scheduled leader, and the Jito block engine auctions atomic bundles that the leader inserts at the top of its block for tips. Ethereum’s pipeline is more explicit and studied, Solana’s faster and more opaque, and both are moving toward encrypted, TEE-based or enshrined block building.

2. Explain it simply

Section titled “2. Explain it simply”

Analogy

Section titled “Analogy”

Think of a concert venue. On Ethereum, fans (users) shout ticket requests in a public square (the mempool); scalpers (searchers) bundle the best requests for promoters (builders) who fill the whole hall, and promoters bid through a trusted escrow agent (the relay) to the night’s venue owner (the proposer), who accepts the highest sealed bid without seeing the seating plan. On Solana, the owner for the next 1.6 seconds is announced in advance; fans mail requests directly, and one company (Jito) runs a side auction where scalpers tip to seat their groups first, all-or-nothing.

ELI5

Section titled “ELI5”

Whoever makes a block decides the order of transactions inside it, and the order is worth money: being first to buy a cheap token or first to close a bad loan pays. On Ethereum this job is split among several companies: some look for profitable orderings, some build whole blocks, some act as a trusted middleman, and the validator just signs the block that pays the most. On Solana the block maker is known ahead of time, transactions are sent straight to it, and a separate service called Jito lets traders pay tips to have a small group of transactions run first, either all together or not at all.

Step-by-step walkthrough

Section titled “Step-by-step walkthrough”

Alice swaps 100 ETH (or the SOL equivalent) for USDC on a DEX, moving the pool price; a searcher wants to backrun her trade with an arbitrage against another pool.

  1. Before. Pool A quotes 3,000 USDC/ETH; pool B quotes 3,000. Alice’s swap pushes pool A to 2,970. Searcher balance: 10 ETH.
  2. Ethereum, discovery. Alice’s transaction sits in the public mempool (or a private order-flow channel). The searcher simulates a bundle [Alice's swap, searcher's arb] that buys ETH in A at 2,970 and sells in B at 3,000, and submits it to several builders with a bid.
  3. Ethereum, block building. Builders merge bundles and mempool transactions into full blocks and send each block plus a payment to relays; relays forward a blinded header and bid to the slot’s proposer, who signs the highest bid, and the relay publishes the block. The arb executes right after Alice’s swap: searcher +1.0 ETH gross, most of it paid on to the builder and, via the builder’s bid, the proposer.
  4. Solana, discovery. Alice’s RPC sends her transaction to the current and next two leaders; there is no shared mempool. The searcher reads the executed swap from the just-produced block (or from ShredStream) and submits a bundle with its arbitrage and a tip transfer to the Jito block engine.
  5. Solana, auction and inclusion. Every 50 ms the block engine auctions bundles that touch overlapping accounts and forwards the highest-tipping set to the leader running Jito-Agave, which reserves compute for bundles at the start of its block. The bundle executes atomically; if any instruction fails, no tip is paid. Searcher +1.0 SOL-equivalent gross minus tip.
  6. After. Both pools are back near parity. Value flowed searcher → builder → proposer on Ethereum and searcher → leader and stakers (minus Jito’s cut) on Solana.

Common misconceptions

Section titled “Common misconceptions”
  • Myth: “Solana has no MEV because it has no mempool.” Reality: Arbitrage, liquidations and sandwiches all happen; Jito processed over 3 billion bundles and 3.75 million SOL of tips over 2024 (Helius, as of 2025-01).
  • Myth: “MEV-Boost relays are just message routers.” Reality: They are trusted escrow agents that protect builders from proposers stealing block contents and protect proposers by validating blocks and payments (Paradigm, 2023-04).
  • Myth: “Validators using MEV-Boost earn most of their income from MEV.” Reality: Block rewards rise from about 3% to 13% of a validator’s annual rewards with MEV-Boost; APR moves from 2.93% to 3.24% (Nero_eth, ethresear.ch, data 2023-11 to 2024-06).
  • Myth: “Sandwiching ended on Solana when Jito closed its mempool.” Reality: Private mempools replaced it; one sandwich program earned 65,880 SOL (about $13.4M) in a 30-day window ending 2025-01-05 (Helius).

If you only remember one thing

Section titled “If you only remember one thing”

Ordering rights are auctioned on both chains; Ethereum made the auction explicit and multi-party (searcher, builder, relay, proposer) while Solana leaves it inside the leader’s slot, with Jito’s tips as the visible price.

3. How it works

Section titled “3. How it works”

Ethereum: from mempool to proposer

Section titled “Ethereum: from mempool to proposer”
Ethereum MEV supply chain under MEV-Boost: users, searchers, builders, relays, proposers Users / walletstxs → mempool / OFASearchersbundles + bidsBuildersfull blocks + bidRelaysescrow, validateProposersigns highest header value flows right → left as bids; ordering rights flow left → right. Solana replaces builders/relays with the leader's scheduler + Jito block engine.
The Ethereum MEV supply chain under MEV-Boost. Transactions and bundles flow left to right; payments flow right to left as bids.

Konstantopoulos and Zhang (Paradigm, 2021-03) describe blockspace as a commodity market whose supply is fixed by block size and block time, distorted by priority gas auctions (PGAs): bots bid gas up to the value of an on-chain opportunity, and the revenue is miner (now maximal) extractable value. Their proposed fix, an endpoint where miners “accept bundles of MEV-only transactions”, became Flashbots and, after the Merge, MEV-Boost.

Under MEV-Boost (Paradigm, 2023-04) a builder assembles a block from users, searchers and private order flow and submits it to a relay; the relay validates it, computes the proposer payment and sends a blinded header plus bid to the proposer, who signs the highest bid; the relay then publishes the block. The relay “protects builders from MEV stealing” and protects proposers by validating hundreds of blocks per slot and checking payments. Slots are 12 seconds with an attestation deadline at t=4t = 4 s; because block value “increases monotonically with time”, proposers play timing games, and “honest reorgs” plus proposer boost (40% of attestation weight) exist to punish late blocks. As of 2023-08, nearly 95% of blocks were built via MEV-Boost (Empirical analysis of builders’ behavioral profiles, ethresear.ch); as of 2024-10, “approximately 90% of blocks on Ethereum are delivered through just a handful of relays”, with relay latency of 5–20 ms (Paradigm, How to Remove the Relay). Exclusive order flow and direct builder payments decide most auctions; beaverbuild won a large fraction of blocks in the 2023 sample.

Economically, for a validator the expected block reward with MEV-Boost is

Rboost≈2.6 blocks/yr×0.0380 ETH≈0.099 ETH/yrR_{\text{boost}} \approx 2.6\ \text{blocks/yr}\times 0.0380\ \text{ETH} \approx 0.099\ \text{ETH/yr}

versus about 2.6×0.0076≈0.0202.6\times 0.0076 \approx 0.020 ETH/yr building locally (the source reports 0.0998 and 0.0199 ETH using unrounded block counts; Nero_eth, ethresear.ch, data 2023-11 to 2024-06), lifting APR from 2.93% to 3.24%.

Solana: leader schedule, TPU and Jito

Section titled “Solana: leader schedule, TPU and Jito”
Solana transaction flow: RPC/TPU to the scheduled leader, with the Jito block engine auctioning bundles alongside User txbase + priority feeSearcher bundleatomic, tip to validatorTPU / QUICstake-weighted QoSJito block engineoff-chain bundle auctionLeader (Jito-Agave)scheduler: fee-priority +bundles at slot start No public mempool: txs go straight to the next leaders. ~400 ms slots, 4 slots per leader.
Solana's flow: user transactions reach the scheduled leader's TPU directly, while Jito's block engine auctions atomic bundles that the leader places at the start of its block.

Solana “does not have a native in-protocol mempool” (Helius). The leader schedule is known an epoch ahead, each leader gets four consecutive ~400 ms slots, and RPCs forward signed transactions to the current leader and the next two. Ingress uses QUIC with stake-weighted quality of service (bandwidth reserved for staked connections), and the banking-stage scheduler orders by fee per compute unit under account locks (see /foundations/fee-markets/). When “you are assigned leader of a given block, you have the authority to determine the contents of your assigned block”, so MEV on Solana is “any value derived from adding, removing, or reordering the transactions within a given block”.

Jito is “an out-of-protocol blockspace auction for partial blocks”. Validators run Jito-Agave, a modified client; the Jito relayer holds incoming transactions for 200 ms while the block engine simulates bundles and runs an auction. Bundles are groups of up to 5 transactions executed “sequentially and atomically”, competing on tips to eight tip accounts; auctions run at 50 ms ticks, with bundles whose account locks intersect placed in one auction and ranked by tip per requested CU, and non-intersecting bundles auctioned in parallel (Jito docs, as of 2026-08). The minimum tip is 10,000 lamports; Jito charged 5% of tips (Helius, as of 2025-01) and 6% split between Jito Labs and the DAO (Helius, as of 2025-08). Jito-Agave “blocks are built with the scheduler reserving 3M CUs for transactions routed via Jito for the first 80% of the block” (Helius). Jito-Agave ran 92% of stake by the start of 2025, up from 48% a year earlier (Helius). Jito’s MempoolStream, a 200 ms preview of pending transactions, was suspended in March 2024 because it was mostly used for sandwiching; DontFront (an account address starting with jitodontfront) makes the block engine keep your transaction first in any bundle. BAM (2025) moves scheduling into AMD SEV-SNP enclaves with attested ordering, “moving Solana closer to a Proposer-Builder Separation (PBS) like architecture” (Helius).

4. Worked numeric example

Section titled “4. Worked numeric example”

Same backrun as §2, with a 1.0 unit gross arbitrage.

Ethereum. The searcher nets 1.0 ETH gross and, under competition, bids 0.90 ETH to the builder; the builder adds 0.20 ETH of other bundles and bids 1.00 ETH to the proposer through the relay. Illustrative split: searcher 0.10, builder 0.10, proposer 1.00. The sourced anchors are that MEV-Boost raises the median block reward from 0.0076 to 0.0380 ETH (ethresear.ch, data to 2024-06) and that PGA-inflated gas already gave miners about 12% of MEV indirectly as of 2021-03 (Paradigm).

Solana. The searcher’s bundle nets 1.0 SOL and attaches a 0.20 SOL tip (far above the 10,000-lamport minimum). Jito’s 5% fee is 0.01 SOL, so 0.19 SOL reaches the leader’s tip distribution (validator commission plus stakers via TipRouter); the searcher keeps 0.80 SOL plus pays the normal per-signature fees. If the arbitrage is not profitable because a faster bundle won the 50 ms auction, the bundle reverts atomically and the tip is not paid, which is why Jito advises putting the tip instruction inside the main transaction rather than a separate one (“uncle bandit” risk).

Sandwich example, real numbers (Helius, 2024-12-16). Front-run: 14.63 SOL buys 32.9M Komeko tokens; the victim fills at the worst price its slippage allowed; back-run: sell for 14.65 SOL, gross 0.02 SOL. The same program averaged 0.0425 SOL ($8.67) per sandwich, 51,600 a day, and 65,880 SOL ($13.43M) profit over 2024-12-07 to 2025-01-05 after 22,760 SOL ($4.63M) of tips (Helius, as of 2025-01).

5. Where it’s used

Section titled “5. Where it’s used”

Ethereum

Section titled “Ethereum”
  • Flashbots Protect / MEV-Share — private submission and order-flow auctions that rebate backrun value — /mev/order-flow-auctions/.
  • CoW Protocol batch auctions — solvers compete off-chain, removing intra-batch ordering MEV — /exchange/batch-auctions/.
  • Builders (beaverbuild and peers) — win blocks through exclusive order flow (ethresear.ch, 2023) — /mev/pbs-and-epbs/.

Solana

Section titled “Solana”
  • Jito bundles and tips — atomic execution and revert protection for bots and searchers; TipRouter distributes tips to stakers — /staking/lrt-risks/.
  • JupiterZ and Kamino Swap RFQ — price discovery off-chain, only the fill on-chain (Helius, 2025-01) — /exchange/rfq-and-intents/.
  • Paladin, sandwich-resistant AMMs, Marinade SAM — validator- and AMM-side mitigations, plus a stake auction critics say rewards sandwiching validators (Helius, 2025-01) — /mev/solana-mev/.

6. Risks, attacks, and incidents

Section titled “6. Risks, attacks, and incidents”
  • Relay unbundling attack, 2023-04-02. A proposer returned an invalid signed header to a buggy relay, obtained block contents and stole about $20M from a searcher; the five patches that followed added latency, and with honest reorgs active up to 13 blocks (4.3%) an hour were reorged on 2023-04-06, about 5x normal (Paradigm, 2023-04).
  • Relay centralisation. About 90% of blocks pass through a handful of relays, and builders co-locate with relays rather than the validator set (Paradigm, 2024-10).
  • Sandwiching. About $24M of sandwich profit on Ethereum in one 30-day window per EigenPhi (via Helius, as of 2024-03). On Solana, DeezNode’s private mempool and its vpeNAL… program account for “nearly half of all sandwich attacks” (Jito via Helius, 2025-01); its validator’s stake rose from 307.9k to 802.5k SOL between epochs 697 and 709 (2024-11 to 2024-12).
  • Stake-centralisation feedback loop. Temporal (2024-12) argued sandwiching validators can outbid others for stake; Marinade’s SAM auction showed winning bids of 13.73% APY against about 9.4% for ordinary validators (Helius, as of epoch 717).
  • Spam as MEV strategy. With low fees and non-deterministic ordering, “over 50% of Solana transactions are on failed arbitrages” (Helius, as of 2024-03); reverts peaked at 75.7% in April 2024.

7. Open problems

Section titled “7. Open problems”
  • Removing relays. Noyes and Policharla propose threshold-encrypting blocks to the attester committee so no trusted relay is needed; adoption hinges on MaxEB and opt-in usage (Paradigm, 2024-10). Enshrined PBS is slated for Glamsterdam (ethereum.org, accessed 2026-08).
  • Timing games. Whether to move the attestation deadline from t=4t=4 to t=6t=6 and how much reorg rate is acceptable is unresolved (Paradigm, 2023-04).
  • Builder concentration. A few builders win most blocks through private flow; inclusion lists (FOCIL) are the proposed counterweight — /mev/mev-burn-tickets-focil-timing/.
  • Private mempools on Solana. Helius warned that “someone else may fork or try to otherwise replicate Jito’s out-of-protocol mempool”, which DeezNode did; validator whitelists are “a measure of last resort”.
  • Trust in Jito and BAM. The block engine is “closed-source and run by a single trusted party”; BAM’s attestations rely on AMD SEV-SNP (Helius, 2025).
  • Alpenglow and multiple leaders. Collapsing the 500–600 ms confirmation window to ~150 ms may favour leaders with custom block-building infrastructure, and concurrent leaders would enable unanalysed cross-lane strategies (Helius, 2025).

8. Ethereum vs Solana

Section titled “8. Ethereum vs Solana”
AspectEthereum (MEV-Boost)Solana (Jito)
Transaction visibilityPublic mempool (plus private OFAs)No in-protocol mempool; direct to leader
Who ordersBuilder (full block)Leader’s scheduler + Jito bundles at top of block
Trusted intermediaryRelay (escrow, validation)Jito block engine (closed-source; BAM in TEEs)
Unit auctionedWhole block, sealed bidBundle ≤5 txs, tip per CU, 50 ms auctions
Producer share of MEVBid to proposer (block reward 0.0076 → 0.0380 ETH median)Tips to leader and stakers (3.75M SOL in 2024)
Adoption~95% of blocks (2023-08)~92% of stake on Jito-Agave (2025-01)
Latency window12 s slot, bids until ~t=0–4 s~400 ms slot, 200 ms relayer delay
Sandwich defencePrivate RPCs, OFAsDontFront, RFQ, sr-AMMs
RoadmapePBS (Glamsterdam), relay removal, FOCILBAM, Alpenglow, MCL

Ethereum made the auction explicit and outsourced, accepting relays as a trusted stopgap; Solana kept ordering inside a known leader’s slot and let one off-chain auctioneer emerge. Both now converge on encrypted or attested block building and protocol-level ordering commitments to shrink the trust placed in intermediaries.

9. Reference doc

Section titled “9. Reference doc”

The reference

Section titled “The reference”

Ethereum Blockspace - Who Gets What and Why — Georgios Konstantopoulos and Leo Zhang (Paradigm), 2021-03-25. https://www.paradigm.xyz/2021/03/ethereum-blockspace-who-gets-what-and-why

Summary of the reference

Section titled “Summary of the reference”

Borrowing Alvin Roth’s market-design lens (a market needs depth, safety and lack of congestion), the post treats blockspace as the commodity that “powers the heartbeats of all cryptocurrency networks”. Supply side: Ethereum mining in 2021 was 80–90% GPU rather than ASIC, giving more OEMs, shorter hardware reaction times and higher option value, hence shorter mining cycles; revenue came from the 2 ETH coinbase, transaction fees and MEV. A key point: more miners raise the security budget but “the supply of blockspace is determined by block size and the average block time”, so hashpower does not lower fees. Demand side: users bid in a repeated first-price auction via gas price; over 75% of pools simply sort by fee; blockspace has time value because unconfirmed transactions face volatility and front-running. Congestion and fees: fee estimation is hard (median gas swung 100–400 gwei in a week), and the three paths are bigger blocks (a painkiller), a new auction (EIP-1559, then scheduled for July 2021 and opposed by over 60% of pool hashpower) and rollups. Dark forest and dark pools: PGAs from Flash Boys 2.0 generate MEV; at least $350M was extracted since 2020 with a third in February 2021, mostly AMM arbitrage; bots in the same pool as users inflate everyone’s fee estimates, so the authors endorse a separate bundle endpoint, “the approach that Flashbots is taking”. How miners adapt: MEV shifts blockspace from a commodity to a matching market; miners already capture ~12% indirectly; time-bandit reorgs are theorised but implausible for ETH-long miners; outsourcing ordering to traders is the optimistic path. Hashpower instruments let DeFi users buy forward MEV exposure. The conclusion cites Charlie Noyes: attempts to block miners from a revenue stream create off-protocol markets.

Key quotes

Section titled “Key quotes”

“In the blockspace market, miners are the producers, mining pools are the auctioneers, and users are the bidders.” — Overview of the blockspace market “Ideally, MEV-transactions should be in a separate transaction pool from non-MEV transactions.” — The Dark Forest and the Dark Pools “All else equal, MEV causes gas prices to be higher than they would be if MEV did not exist.” — How will miners adapt to MEV? “Any attempt to prevent miners from accessing the revenue stream is liable to incentivize the creation of off-protocol markets.” — Conclusion, quoting Charlie Noyes

How to read the original

Section titled “How to read the original”

Background: what a block and a mempool are, and first-price auctions. Read the Overview, “Congestion and fees” and “The Dark Forest and the Dark Pools” first; skip the GPU-vs-ASIC supply chain on a first pass unless you care about mining economics. The hardest idea is that MEV is a matching-market problem layered on a commodity market: once ordering has value, whom you send your transaction to matters, which is the seed of the entire searcher-builder-relay pipeline.

What changed since

Section titled “What changed since”

EIP-1559 shipped (2021-08) and the Merge (2022-09) replaced miners with proposers; Flashbots’ bundle endpoint became MEV-Boost with builders and relays, carrying ~95% of blocks by 2023-08. The 2023-04 relay exploit exposed relay fragility (Paradigm, 2023-04), and by 2024-10 Paradigm was proposing to remove relays altogether. Ethereum’s roadmap now enshrines PBS in Glamsterdam (planned Q4 2026). Hashpower instruments became moot after the Merge; MEV exposure is now bought through staking and restaking products.

Secondary references

Section titled “Secondary references”
  1. Paradigm, “Time, slots, and the ordering of events in Ethereum Proof-of-Stake” (2023-04) — read if you want the MEV-Boost step sequence and the relay incident.
  2. Paradigm, “How to Remove the Relay” (2024-10) — read if you want the threshold-encryption alternative.
  3. Nero_eth, “Is it worth using MEV-Boost?” (ethresear.ch, 2024-06) — read for validator-level numbers.
  4. “Empirical analysis of Builders’ Behavioral Profiles” (ethresear.ch, 2023-08) — read for builder concentration data.

The reference

Section titled “The reference”

Solana MEV: An Introduction — Helius (with review from Overclock, Jito and Ellipsis), 2024. https://www.helius.dev/blog/solana-mev-an-introduction

Summary of the reference

Section titled “Summary of the reference”

Takeaways: MEV on Solana is not dead, not all MEV is bad, front-running is possible on any venue, continuous block production without a mempool changes the social dynamics, replicating Jito’s mempool is hard, and many validators supported removing it. Introduction: the leader controls block contents; a searcher tipped 890 SOL on 2024-01-10 and validators earned over $7M of tips in the week to 2024-03-12; over 50% of transactions were failed arbitrages. Structure: searchers co-locate with staked nodes because Turbine and stake-weighted QoS are latency- and stake-sensitive; there is no conventional mempool, which reduces front-running surface. MEV types: NFT minting (source of 2021–22 congestion halts), liquidations, arbitrage (only intra-chain arbitrage is atomic). Jito: an out-of-protocol auction for partial blocks; bundles are atomic; minimum tip 10,000 lamports; tips are separate from priority fees; auctions run off-chain so only winners hit the block. Mempool: Jito’s now-deprecated 200 ms pseudo-mempool became canonical because ~65% of validators ran Jito-Solana; sandwiching was a large share of tip revenue and users were filled at their max slippage. Supply chain: incoming transactions → relayers (dedup, sigverify, forward to block engine and validator) → block engine (simulates, auctions) → validators, with 3M CU reserved for Jito traffic in the first 80% of the block; entities may be vertically integrated. Distribution: latency-driven centralisation risk under sub-200 ms blocks. Reducing surface: RFQ systems and MEV-protected RPCs with rebates, each trading off trustlessness.

Key quotes

Section titled “Key quotes”

“Jito is an out-of-protocol blockspace auction for partial blocks, unlike MEV-boost where full blocks are built” — Jito “Unlike Ethereum, Solana does not have a native in-protocol mempool.” — Mempool “Jito-Solana blocks are built with the scheduler reserving 3M CUs for transactions routed via Jito for the first 80% of the block.” — Supply Chain “Someone else may fork or try to otherwise replicate Jito’s out-of-protocol mempool to extract more MEV, but it is hard both technically and socially.” — Takeaways

How to read the original

Section titled “How to read the original”

Background: leader schedules and what a bundle is. Read Takeaways, Jito, Mempool and Supply Chain; skip the RFQ section if you already know off-chain quoting. The hardest paragraph is the relayer description: Solana relayers are transaction proxies doing deduplication and signature verification, not the trusted escrow role Ethereum relays play, and the 200 ms hold is what creates the auction window.

What changed since

Section titled “What changed since”

Helius’ MEV Report (2025-01) quantified the year after the mempool shutdown: 3 billion bundles, 3.75M SOL tips, 90.4M arbitrages averaging $1.58, Jito-Agave at 92% of stake, and DeezNode’s private mempool responsible for nearly half of sandwiches. Agave v1.18 (2024-05) made ordering more deterministic. Jito shipped DontFront and, in 2025, BAM with TEE-based scheduling and a stated PBS-like separation; the Anza roadmap (2025-07) places application-controlled execution, Alpenglow and MCL as the next steps.

Secondary references

Section titled “Secondary references”
  1. Helius, “Solana MEV Report” (2025-01) — read for the data and the DeezNode case study.
  2. Jito Labs docs, “Low Latency Transaction Send” — read for bundle, auction and tip mechanics.
  3. Solana docs, “MEV Protection with Jito DontFront” — read if you are integrating sandwich protection.
  4. Helius, “Block Assembly Marketplace (BAM)” (2025) — read for where Solana block building is heading.

10. Sources

Section titled “10. Sources”