Solana MEV

1. TL;DR

Solana has no protocol mempool: transactions are forwarded straight to the next scheduled leaders, so the classic public-mempool sandwich has no natural home, and most MEV is latency-driven arbitrage plus backruns landed through Jito’s out-of-protocol bundle auction. Jito’s block engine became the de facto MEV market (92% of stake ran Jito-Solana by January 2025), collected 3.75 million SOL of tips in 2024, and suspended its own 200 ms pseudo-mempool in March 2024 because it enabled sandwiching; sandwiching then moved to private validator mempools and still cost users over $7.7M on Jito bundles in four months of 2025. The next step is BAM (2025), which moves sequencing into TEEs with app-defined ordering plugins.

2. Explain it simply

Analogy

Ethereum is a post office with a public sorting room where everyone can see the letters before they are delivered. Solana is a courier service: your letter goes directly to the driver whose shift is next, so strangers cannot read it in the sorting room. But the driver can, and a company (Jito) sells drivers a side-business: anyone can pay a tip to have a small stack of letters delivered together, in a fixed order, right at the start of the shift. For a while Jito also let tippers peek at everyone’s letters for a fifth of a second; it stopped when the peeking was used to rob senders.

ELI5

On Solana, your transaction is sent directly to the validator whose turn is next, not to a public waiting room, so bots cannot easily see it and jump in front. Instead, bots race on speed: they watch what just happened and send their own trade as fast as possible, or they pay a tip to a service called Jito to land a small group of trades together in a guaranteed order. Because sending is cheap, bots also spam thousands of trades hoping some hit; at one point most transactions on the network were failed bot trades. Some validators secretly recreated the waiting room to squeeze users; the community pushed back by cutting their stake.

Step-by-step walkthrough

The 10 January 2024 WIF backrun (all figures from Helius’s report):

1. Before: WIF trades near $0.20 with a few million dollars of liquidity across three pools. A user submits a single Jupiter swap buying $8.9 million of WIF; a searcher’s node sees the swap land in the leader’s block.
2. Opportunity: the buy drains the three pools, pushing WIF’s on-chain price far above every other venue. Whoever trades next can buy WIF cheaply elsewhere and sell into the distorted pools.
3. Bundle: the searcher submits a Jito bundle: (a) swap 703.31 SOL ($72,368) for 490,143.90 WIF in a Raydium concentrated-liquidity pool; (b) sell those WIF into the drained pools for 19,035.97 SOL ($1,958,733); (c) a tip transfer of 890.42 SOL (~$91,621) to a Jito tip account.
4. Auction: the Jito block engine simulates competing bundles touching the same accounts and forwards the highest tip-per-compute combination to the current Jito-Solana leader, which executes it atomically at the start of its block.
5. After: the searcher’s SOL balance rises by $19{,}035.97 - 703.31 - 890.42 = 17{,}442.24$ SOL (~$1.79M); the leader and its stakers receive the 890.42 SOL tip (minus Jito’s fee); the user who bought WIF is unaffected by the backrun itself but has already paid the price impact of routing through thin pools.

Common misconceptions

• Myth: No mempool means no MEV. Reality: Jito’s arbitrage detector counted 90,445,905 successful arbitrages in 2024 worth $142.8M (as of 2025-01); NFT-mint sniping, liquidations, and backruns all exist.
• Myth: Jito is Solana’s MEV-Boost. Reality: MEV-Boost auctions whole blocks built by builders; Jito auctions bundles (≤5 transactions) that the leader inserts into its own block; the leader still builds and executes everything.
• Myth: The March 2024 mempool shutdown ended sandwiching. Reality: it removed the public 200 ms window, but private mempools (DeezNode and others) and Jito bundles themselves still enable it; Gerzon et al. found 521,903 sandwiches in Feb–Jun 2025 (as of 2025-06).
• Myth: Priority fees buy a guaranteed position. Reality: Solana’s scheduler is not strict priority ordering; that is why spamming duplicate transactions was rational and why Paradigm’s MEV taxes “would likely not work” here without BAM-style enforcement.
• Myth: Jito tips are a small side income. Reality: 3.75 million SOL of tips flowed to validators in 2024, peaking at 60,801 SOL in a single day (as of 2025-01).

If you only remember one thing

Solana replaces the public-mempool race with a latency race and a tip auction; its MEV problems are about who gets to see flow (Jito, private mempools, TEEs), not about who gets to build the block.

3. How it works

Solana's two lanes into a Jito-Solana leader: ordinary transactions via the TPU with stake-weighted QoS, and tipped bundles via the Jito block engine's off-chain auction.

No mempool, continuous leaders

A leader is scheduled by stake-weighted lottery an epoch in advance and produces four consecutive ~400 ms slots. Clients forward transactions over QUIC to the current and upcoming leaders’ transaction processing units (TPUs); stake-weighted quality of service reserves capacity for staked senders; unprocessed transactions expire after ~150 blocks (~1 minute). Continuous block production “reduces the surface area for certain types of MEV (particularly front-running)” (Helius), so the incentive shifts to latency: searchers “co-locate with high-staked nodes for the most up-to-date view of the chain”.

Spam as a strategy

With tiny fees and non-deterministic scheduling, spamming probabilistic arbitrage attempts was positive expected value: “over 50% of Solana transactions are on failed arbitrages (spam)” (as of 2024-03); reverted transactions peaked at 75.7% of non-vote transactions in April 2024 before the Agave 1.18 central scheduler (as of 2025-01). Jito estimated validators spent ~58% of their time on failed arbitrages (cited in ethresear.ch/19695).

Jito: bundles, tips, block engine

Jito-Solana is a fork of the Agave client. Searchers, apps, and Telegram bots send bundles of up to five transactions to a block engine; bundles execute “sequentially and atomically”, all-or-nothing, within one slot, and must tip one of eight tip accounts (minimum 1,000 lamports per current docs; 10,000 in Helius’s 2024 description). The engine runs “parallel auctions … at 50ms ticks”: bundles whose account locks intersect compete in one auction, ranked by tip per compute unit; the highest-paying combination goes to the leader “up to some CU limit”, which the Jito-Solana scheduler reserves at the start of the block. Jito relayers deduplicate, forward, and briefly delay packets so the auction can run; they are not escrow relays. Jito takes a 6% fee on tips (DAO revenue of 22,391.31 SOL ≈ $4M in Q2 2025); the rest flows to the leader and its stakers.

Formally, for bundle $i$ with tip $\tau_i$ and requested compute $c_i$, the engine ranks conflicting bundles by $\tau_i / c_i$ and selects the feasible set maximizing $\sum \tau_i$ subject to $\sum c_i \le C_{\text{bundle}}$, the reserved compute budget.

The mempool episode (2022–March 2024)

Jito’s block engine initially exposed a MempoolStream: transactions sat in a pseudo-mempool for 200 ms while searchers bid to front-run, sandwich, or backrun them. With ~65% of validators on Jito-Solana (as of 2024-03) this was “effectively a canonical out-of-protocol mempool”, and “sandwiching represented a significant portion of MEV income”. On 8 March 2024 Jito Labs suspended it “due to negative externalities impacting users on Solana” (Jito Labs on X), “sacrificing significant revenue”; bundles continued.

Protections: DontFront, dynamic slippage, MEV-protect

DontFront: a transaction referencing any account whose address starts with jitodontfront must be at index 0 of any bundle containing it, so it can be backrun but never frontrun; violating bundles are rejected. Jupiter dynamic slippage (August 2024) computes per-trade tolerance heuristically. MEV Protect mode routes only to Jito block engines. Sandwich-resistant AMMs (Ellipsis’s Plasma, November 2024) forbid executing at a better price than the pool’s price at the start of the slot window. Paladin patches Jito-Solana to drop sandwich bundles by pattern (80 validators, 6% of stake, as of 2025-01).

BAM: TEE sequencing with plugins

Announced July 2025, BAM splits sequencing (BAM nodes, running a TPU and scheduler inside AMD SEV-SNP enclaves) from execution (validators, which must schedule the received sequence exactly). Transactions are encrypted until execution; nodes sign attestations of the sequence, producing an audit trail. Plugins implement application-controlled execution (cancel-before-take for order books, just-in-time oracle updates); the first scheduler runs periodic intra-block auctions subdividing the block into $N$ sub-auctions. Helius reads this as Solana moving “closer to a Proposer-Builder Separation (PBS) like architecture” (see /mev/pbs-and-epbs/ and /mev/encrypted-mempools/).

4. Worked numeric example

The WIF backrun (§2), in SOL. Cost of the front leg: 703.31 SOL. Proceeds of the back leg: 19,035.97 SOL. Tip: 890.42 SOL. Net $= 19{,}035.97 - 703.31 - 890.42 = 17{,}442.24$ SOL. At the implied price of the tip ($91,621 / 890.42 = $102.9/SOL) that is about $1.79M. The tip was $890.42 / 18{,}332.66 = 4.9\%$ of gross profit: the searcher faced little competition for a one-off opportunity, so the auction cleared far below the opportunity’s value. In a competitive Jito auction the tip converges toward the second-highest bidder’s valuation, exactly as in MEV-Boost.

A Komeko sandwich (16 December 2024, Helius). Front-run: 14.63 SOL buys 32.9 million KOMEKO on a Pump.fun-launched pool. Victim: buys, pushing the price up. Back-run: 32.9 million KOMEKO sold for 14.65 SOL. Tip: 0.000148 SOL ($0.03). Net $\approx 14.65 - 14.63 - 0.000148 \approx 0.02$ SOL; from unrounded amounts Helius reports 0.01678 SOL ($3.35). Multiply by the vpe program’s ~51,600 sandwiches per day at an 88.9% success rate and you get its measured ~2,200 SOL/day (as of 2025-01). Tips paid over that month, 22,760 SOL ($4.63M), were 34.5% of its 65,880 SOL profit, a much higher tip share than the WIF backrun because sandwich bundles compete every slot.

Defensive bundling. Over 86% of single-transaction Jito bundles carried tips too small to buy priority; users spent $2,421,868 on them in Feb–Jun 2025, $0.00286 per bundle, to be un-sandwichable within Jito (Gerzon et al.). Against a median loss of ~$5 and a 0.038% sandwich rate, the expected loss avoided is $\text{\textdollar}5 \times 0.00038 \approx \text{\textdollar}0.0019$ per bundle, below the tip: fear of loss, not expected loss, drives the behaviour.

5. Where it’s used

Ethereum

• MEV-Boost — the closest analogue to Jito, but for full blocks; Ethereum’s public mempool is why sandwiches there historically dominated toxic MEV. See /mev/pbs-and-epbs/.
• Flashbots Protect / MEV-Share — refund-based protection; Solana’s protect modes protect but rarely refund. See /mev/order-flow-auctions/.

Solana

• Jito block engine and Jito-Solana — 48% of stake at the start of 2024, 92% (373.8M SOL) by January 2025; 3 billion bundles and 3.75 million SOL in tips in 2024; daily tippers from ~20,000 to ~938,000 (as of 2025-01). docs.jito.wtf
• Jupiter — dynamic slippage (August 2024) and MEV-protect routing to Jito; see /exchange/routing-and-aggregation/.
• Telegram trading bots (BonkBot, Trojan, Photon) — the main source of high-slippage memecoin flow that sandwichers target; 16 of the top 20 sandwiched tokens were Pump.fun mints (as of 2025-01).
• BAM — permissioned launch with Helius, SOL Strategies, Triton One, Figment; plugin partners Drift, Pyth, DFlow (as of 2025-07).

6. Risks, attacks, and incidents

• Jito mempool shutdown (8 March 2024). Jito Labs suspended MempoolStream “due to negative externalities impacting users”; Helius calls it a “contentious decision” that cost Jito significant revenue and immediately reduced sandwiching but “encouraged the rise of alternative mempools that lack transparency”.
• DeezNode private mempool (2024–2025). Validators reported receiving a “DeezMempool” proposal with profit-share terms shortly after the shutdown. Jito’s internal analysis attributed nearly half of all Solana sandwiches to the vpe program: 1.55 million sandwich transactions from 7 December 2024 to 5 January 2025, 88.9% success, 65,880 SOL ($13.43M) profit, 22,760 SOL of tips; the operator’s validator held 811,604.73 SOL ($168.5M) of delegated stake (as of 2025-01, Helius).
• Solana Foundation delegation removals (June 2024). Over 30 operators were removed from the Foundation Delegation Program for participating in sandwiching mempools; enforcement continued into 2025 (The Block, secondary; Gerzon et al. note the Foundation “banned a number of such validators”).
• Measured sandwich losses on Jito (Feb–Jun 2025). 521,903 sandwiches; victims lost over $7.7M, attackers gained $9,678,466; median loss ~$5 with a tail over $100; sandwich bundles carried median tips over 2,000,000 lamports versus 1,000 for ordinary three-transaction bundles (Gerzon et al., as of 2025-06).
• Marinade SAM and pay-for-stake. SAM’s winning bids (13.73% APY, top ten 18.27%) exceed what honest validators can pay (~9.4%); Helius links the gap to sandwich revenue (as of 2024-12).
• Trust in Jito. The block engine is closed-source and run by one company; BAM promises open-sourcing and attestations but launched permissioned (as of 2025-07).

7. Open problems

• Who should be allowed to see flow? Helius argues public mempools “promote a fairer distribution of MEV opportunities” than private permissioned ones, yet the public one was shut down; BAM’s encrypted mempool is the proposed synthesis.
• Enforcing ordering rules. Without a builder market, Solana cannot enforce “competitive priority ordering”; Robinson and White’s MEV taxes therefore do not apply, and BAM’s attested sequences are the first credible enforcement (as of 2025-07).
• Multiple concurrent leaders and asynchronous execution. Helius lists both as protocol-level ways to shrink the sandwich window; neither is scheduled (as of 2025-01).
• Measurement. Jito data “does not capture the full scope of MEV activity”; Gerzon et al. had to reverse-engineer an undocumented Jito Explorer API.
• Redistribution vs relocation. BAM channels MEV to plugin authors, node operators, validators, and the DAO; Helius warns “BAM does not erase MEV—it relocates it”.

8. Ethereum vs Solana

Aspect	Ethereum	Solana
Pending-tx visibility	public mempool (private share >50% of gas in 2025)	none in protocol; Jito engine, private validator mempools
Block producer	proposer signs a builder’s block blind	leader builds and executes its own 4 slots
MEV auction unit	whole block (MEV-Boost)	bundle ≤5 txs (Jito), 50 ms auctions
Toxic MEV share	sandwiches ~$2.5M/month (Oct 2025, secondary)	0.038% of Jito bundles are sandwiches; >$7.7M victim loss Feb–Jun 2025
Benign MEV	CEX-DEX and DEX-DEX arbitrage	90.4M atomic arbitrages, $142.8M profit (2024)
User protection	Protect/MEV Blocker refunds ~90%	DontFront, MEV-protect routing, dynamic slippage; no refund norm
Next architecture	ePBS (EIP-7732), FOCIL	BAM (TEE sequencing + plugins)

Ethereum’s MEV market is a building market layered on a visible mempool; Solana’s is a latency and access market layered on a leader that already sees everything. That is why Ethereum’s reforms target the proposer–builder relationship while Solana’s target who may observe transactions and whether the leader can be made to prove it followed a sequence. Both are converging on TEE-attested sequencing.

9. Reference doc

The reference

Solana MEV Report: Trends, Insights, and Challenges — Helius (Lostin), 2025-01. https://www.helius.dev/blog/solana-mev-report

Summary of the reference

The report opens with key takeaways: MEV on Solana differs because there is no global mempool, so out-of-protocol mempools require adoption by a large share of stake; Jito suspended its public mempool in March 2024 at significant revenue cost, which reduced harmful MEV but encouraged opaque private mempools; memecoin traders using Telegram bots with high slippage are the main sandwich victims; Marinade’s SAM lets sandwiching validators buy stake; much of Solana’s sandwiching comes from DeezNode’s private mempool; Jito bundles are the primary ordering tool but Jito data undercounts MEV; and sandwich-resistant AMMs are an emerging design.

A “History” section walks through: August 2022, the Jito-Solana client launch and its block engine/relayer architecture (adoption under 10% for nine months, 50% by January 2024, over 92% by stake today); January 2024, memecoin season and the shift to Telegram bots; March 2024, the mempool suspension; May 2024, the Agave 1.18 central scheduler; June 2024, Marinade’s SAM (winning APY 13.73%, top-ten 18.27% versus ~9.4% for zero-commission validators at epoch 717); December 2024, concern over DeezNode’s validator (811,604.73 SOL, ~$168.5M) and Jito’s finding of sandwiches outside its auction.

“Types of MEV” gives worked examples: an Orca–Phoenix arbitrage netting 0.00013 SOL-equivalent; front-running; the 10 January 2024 WIF backrun (703.31 SOL → 490,143.90 WIF → 19,035.97 SOL with an 890.42 SOL tip); and a 16 December 2024 Komeko sandwich (14.63 SOL in, 14.65 SOL out, 0.000148 SOL tip, 0.01678 SOL profit).

“Current landscape” quantifies 2024: over 3 billion bundles and 3.75 million SOL in tips (781 SOL on the lowest day, 60,801 SOL on 19 November); bundles peaking at 24.4 million on 21 December; daily tippers from ~20,000 to ~938,000; Jito-Solana stake from 189.5M SOL (48%) to 373.8M SOL (92%); reverted transactions peaking at 75.7% in April; 90,445,905 arbitrages averaging $1.58 for $142.8M, 88.7% in SOL. The vpe case study: 1.55 million sandwiches in 30 days, 88.9% success, 65,880 SOL profit, 22,760 SOL tips, mostly Raydium victims and Pump.fun tokens; annualized 801,540 SOL, implying +0.2% stake share if reinvested.

“Mitigations” covers social ostracism (rejected as a last resort), Jupiter dynamic slippage, MEV Protect mode, sandwich-resistant AMMs (Plasma), DFlow conditional liquidity and declarative swaps, Paladin (80 validators, 6% of stake, claimed +12.5% block rewards), multiple concurrent leaders, and asynchronous execution.

Key quotes

“Jito suspended its public mempool in March 2024, incurring a significant revenue loss. This move immediately reduced harmful MEV practices.” (§Key takeaways)

“Over the past year, more than 3 billion Jito bundles were processed, generating 3.75 million SOL in total tips.” (§Key takeaways)

“Internal analysis from Jito indicates that nearly half of all sandwich attacks on Solana can be attributed to this single program.” (§Case Study: Vpe Sandwich Program)

“Reverted transactions peaked in April 2024, accounting for 75.7% of all non-vote transactions.” (§Current landscape)

How to read the original

Background: Solana’s leader schedule, compute units and priority fees, and what a Jito bundle is (Helius’s earlier “Solana MEV: An Introduction” covers these). Read the takeaways, the history timeline, and the vpe case study; skip the mitigation catalogue on first pass. The hardest part is interpreting Jito tip data: tips measure demand for guaranteed inclusion, not MEV profit, and exclude private-mempool activity, so they are simultaneously an overcount (non-MEV users tipping for landing) and an undercount (sandwiches outside Jito).

What changed since

• 2025-02 to 2025-06: Gerzon et al. measured 521,903 sandwiches on Jito bundles with >$7.7M in losses and $2.4M spent on defensive bundling.
• 2025-07: Jito announced BAM (TEE sequencing, plugins, DAO fee redirection); Jito-Agave client adoption at 79% of validators (Helius, as of 2025-07).
• 2025-09: 97% of the top 500 validators ran a Jito-compatible client (Gerzon et al.).
• 2025: Solana’s jitodontfront guide became part of official Solana developer documentation.

Secondary references

1. “Solana MEV: An Introduction” (Helius, 2024) — read first if you need the supply-chain diagram and the mempool-era context.
2. “Quantifying the Threat of Sandwiching MEV on Jito” (Gerzon et al., IMC ‘25) — read if you want an independent measurement with methodology.
3. “Block Assembly Marketplace (BAM)” (Helius, 2025) — read if you want where Solana block building is going.
4. Jito “Low Latency Transaction Send” docs — read if you want the exact bundle, tip, and auction rules.
5. “Priority Is All You Need” (Paradigm, 2024) — read if you want to understand why Solana’s ordering cannot yet support MEV taxes.

The reference

Quantifying the Threat of Sandwiching MEV on Jito: A Measurement of Solana’s Leading Validator Client — Nicole Gerzon, Ben Weintraub, Junbeom In, Alan Mislove, Cristina Nita-Rotaru (Northeastern University), ACM IMC ‘25, 2025-10. https://doi.org/10.1145/3730567.3764493

Summary of the reference

§1 motivates the study: Solana’s design “lacks a public mempool … making MEV impossible for non-validator users”, but Jito’s August 2022 client opened one; Jito suspended it in March 2024, yet tips and utilization only increased and validators reportedly recreated mempools privately. Contributions: a methodology to collect Jito bundle data, analysis of four months of 2025 data finding 521,903 sandwiches costing users over $7.7M, and detection of defensive bundling. §2 gives background on Solana (400 ms blocks, 5,000-lamport base fee, ledger ~400 TB as of April 2025), sandwiching (Table 1 example), and Jito (97% of the top 500 validators, as of September 2025). §3 describes collecting bundles from 9 February to 9 June 2025 via an undocumented Jito Explorer API returning up to 50,000 bundles per request (95% overlap between successive calls suggests near-complete coverage; 14.8M bundles and 26M transactions per day on average), the sandwich heuristic (first and third transactions by the same account, opposite direction, net profit), and the defensive-bundle classifier (length-one bundles with tips below priority-relevant levels). §4 reports results: sandwiches are 0.038% of bundles but victims lost over $7.7M while attackers gained $9,678,466 (SOL-denominated pairs only); median victim loss ~$5 with some over $100; attacks declined over the period while defensive bundles rose; over 86% (864,889,302) of length-one bundles were defensive, costing $2,421,868 in tips; sandwich bundles’ median tip exceeded 2,000,000 lamports versus 1,000 for ordinary length-three bundles. §5 concludes that widespread anticipation of attacks, despite their rarity, calls for “more transparent governance around validator-driven extensions”.

Key quotes

“finding over 500K instances of Sandwiching attacks resulting in over $7.7M in losses for victims.” (Abstract)

“We collected data from February 9, 2025 until June 9, 2025.” (§3.1)

“over 86% (864,889,302) of the bundles of length one have Jito tips that are too small to serve as a meaningful incentive for prioritization” (§4.2)

“While the median length-three bundle has a Jito tip of 1,000 lamports, the median Sandwiching bundle has a Jito tip of over 2,000,000 lamports” (§4.2)

How to read the original

Background: Jito bundles and tip accounts. Read the abstract, §3.2–3.3 (heuristics), and §4; the data-collection details in §3.1 matter only if you want to reproduce it. The hardest point is the profit/loss accounting: only sandwiches with SOL on one side (72%) can be valued, so the $7.7M is a lower bound over a subset, and SOL/USD is fixed at the 12 September 2025 rate.

What changed since

• 2025-07: BAM announced, targeting exactly the “validator-driven extension” transparency gap the paper raises.
• 2025–2026: DontFront documented in official Solana docs; measurement of BAM-era sandwiching not yet published (as of 2026-08).

Secondary references

1. “Solana MEV Report” (Helius, 2025) — read for the industry-side numbers and the DeezNode case.
2. “MEV Protection with Jito DontFront” (Solana docs) — read for the defensive mechanism the paper’s users approximate with defensive bundling.
3. “Blockchain Transaction Conflicts: A Historical Perspective” (Anjana, Ravi, Herlihy, 2025) — read if you want the concurrency-theory framing of ordering conflicts.

10. Sources

• Solana MEV Report: Trends, Insights, and Challenges — Helius — 2025-01 — https://www.helius.dev/blog/solana-mev-report
• Solana MEV: An Introduction — Helius — n/d (c. 2024-03) — https://www.helius.dev/blog/solana-mev-an-introduction
• Quantifying the Threat of Sandwiching MEV on Jito: A Measurement of Solana’s Leading Validator Client — Gerzon, Weintraub, In, Mislove, Nita-Rotaru — 2025-10 — https://doi.org/10.1145/3730567.3764493
• Block Assembly Marketplace (BAM) — Helius — 2025-07 — https://www.helius.dev/blog/block-assembly-marketplace-bam
• Low Latency Transaction Send — Jito Labs docs — n/d — https://docs.jito.wtf/lowlatencytxnsend/
• What is Jito? — Jito Labs docs — n/d — https://docs.jito.wtf/
• MEV Protection with Jito DontFront — Solana developer docs — n/d — https://solana.com/developers/guides/advanced/mev-protection
• Priority Is All You Need — Dan Robinson, Dave White (Paradigm) — 2024-06-04 — https://www.paradigm.xyz/writing/priority-is-all-you-need
• Strawmanning Based Preconfirmations — Lin Oshitani — 2024-05-31 — https://ethresear.ch/t/strawmanning-based-preconfirmations/19695
• Jito Labs suspension announcement — Jito Labs on X — 2024-03-08 — https://x.com/jito_labs/status/1766228889888514501
• Solana Foundation removes certain operators from delegation program over malicious sandwich attacks — The Block (secondary) — 2024-06-10 — https://www.theblock.co/post/299244/solana-foundation-removes-certain-operators-from-delegation-program-over-malicious-sandwich-attacks
• Exclusive data from EigenPhi reveals that sandwich attacks on Ethereum have waned — Cointelegraph (secondary) — 2025-10 — https://www.tradingview.com/news/cointelegraph:fa12ba092094b:0-exclusive-data-from-eigenphi-reveals-that-sandwich-attacks-on-ethereum-have-waned/
• 2 Million Protect Users — Flashbots — 2025 — https://writings.flashbots.net/2m-protect-users
• Stake-Weighted Quality of Service: Everything You Need to Know — Helius — n/d — https://www.helius.dev/blog/stake-weighted-quality-of-service-everything-you-need-to-know