A Flash(bot) in the Pan: Measuring Maximal Extractable Value in Private Pools

• The study examines Flashbots’ impact and claimed goals, as well as other private pools and their role in MEV extraction.Source: §1 · p. 2
• 97.6% of Ethereum hashing power used Flashbots five months after release, rising to around 99.9% after eight months.Source: §1 · p. 2
• Flashbots benefits miners disproportionately: miners using it make more profit than before, while non-miners make less.The authors also report that some users have left the system.Source: §1 · p. 2
• Other private pools still host MEV extraction, sometimes through pools consisting of a single miner.Source: §1 · p. 2

2.1 Ethereum

• Miners select mempool transactions and control intra-block ordering; transaction execution can depend on preceding transactions.Source: §2.1 · p. 2

2.2 DEXes, Frontrunning, and MEV

• Frontrunning and backrunning place an extractor’s transaction before or after a target, while MEV seeks profit from protocol or market opportunities.Source: §2.2.1 · p. 2, §2.2.2 · p. 2

2.2.2 MEV Extraction.

• Sandwiching surrounds a victim trade with the extractor’s buy or sell before it and a reverse trade after it; arbitrage exploits cross-exchange price differences.Source: §2.2.2 · p. 3
• Liquidators receive discounted collateral for repaying unhealthy loans, and fixed-spread liquidations are atomic, first-come-first-served opportunities for frontrunning.Source: §2.2.2 · p. 3

2.3 Flash Loans

• Flash loans are borrowed and repaid within one transaction, enabling collateral-free capital for arbitrage and liquidation MEV but not sandwiching.Source: §2.3 · p. 4

2.4 Private Transactions

• Private transactions are sent directly to trusted miners rather than publicly propagated, supporting privacy and attempts to avoid or conduct frontrunning.Source: §2.4 · p. 4

2.5 Flashbots

• Flashbots routes searcher bundles through relays to miners; bundles are immutable and atomic, and miners choose which profitable bundles to include.Source: §2.5 · p. 4

2.5.1 Goals of Flashbots.

• Flashbots states goals of illuminating MEV, democratizing extraction through open infrastructure, and distributing MEV benefits across Ethereum participants.Source: §2.5.1 · p. 5

3 DATA COLLECTION AND PROCESSING

• The researchers used an Ethereum archive node and prior detection heuristics, collecting blockchain data from blocks 10,000,000–14,444,725.Source: §3 · p. 5, §3.1 · p. 5

3.1 MEV

• Sandwich detection used token-transfer events and heuristics requiring same-block transactions, near-matching attacker buy and sell amounts, and a higher gas price on the first attacker transaction.Source: §3.1.1 · p. 5
• Among the collected cases, Flashbots were used in 485,680 of 1,020,044 sandwich attacks (47.61%), 916,709 of 3,462,678 arbitrages (26.47%), and 9,191 of 32,819 liquidations (28.01%).Source: §3.1.1 · p. 5, §3.1.2 · p. 6, §3.1.3 · p. 6

3.1.3 Liquidation.

• Liquidation detection used Aave and Compound events, and profit was computed from received collateral minus debt, transaction fees, and applicable miner tips.Source: §3.1.3 · p. 6

3.1.4 Limitations.

• The measurements are lower bounds: sandwich detection misses profitable attacks spanning blocks, excludes MEV types beyond sandwiching, arbitrage, and liquidation, and covers only popular platforms.Source: §3.1.4 · p. 6

3.2 Pending Transactions

• The team recorded 125,660,856 public pending transactions over five months, from November 8, 2021 to April 9, 2022.Source: §3.2 · p. 6

3.3 Flashbots API

• The researchers downloaded Flashbots block records through block 14,444,725 and used included transactions to label Flashbots activity in their analysis.Source: §3.3 · p. 6

3.4 Flash Loan Extraction

• Flash-loan detection relied on lending-platform events for successful loans on Aave and dYdX.Source: §3.4 · p. 6

4.1 Flashbots Bundles

• 3,249,003 bundles appeared in 1,196,218 blocks, averaging 2.71 bundles per block and 2.15 transactions per bundle.61.37% of bundles contained one transaction; the maximum-size bundle contained 700 F2Pool miner payouts.Source: §4.1 · pp. 6–7

4.2 Flashbots Blocks Vs. Non-Flashbots Blocks

• Flashbots blocks peaked at 60.6% of Ethereum blocks in July 2021, then fell to 48.2% by February 2022.The authors interpret the decline as indicating fewer bundles were likely available.Source: Figure 3 · p. 7, §4.2 · p. 7

4.1 Flashbots Bundles

• 90.5% of bundles were Flashbots bundles, 7.6% were marked rogue, and 1.9% were miner payouts.The authors interpret most bundles as submitted either to extract MEV or to protect against MEV extractors.Source: §4.1 · p. 7

4.3 Hashing Power

• By February 2022, Flashbots miners accounted for around 99.9% of Ethereum hashpower, up from 97.6% in May 2021.The authors estimate miner hashpower from the share of network blocks mined; they suspect the Flashbots dashboard's 74.5% estimate is an underestimate because its methodology is unavailable.Source: §4.3 · p. 7, §4.4 · p. 7, Figure 4 · p. 8

4.4 Participating Miners

• At the adoption peak, one or two miners each mined over 10,000 blocks but represented less than 2% of Flashbots miners; no month had more than 55 participating miners.The distribution of blocks mined, and therefore hashpower, was long-tailed and did not appear to indicate democratization of MEV.Source: Figure 5 · p. 8, §4.4 · p. 8

4.5 MEV Usage

• Sandwich transaction counts declined alongside Flashbots usage, and the April 2021 gas-price drop aligned more closely with that usage decline than with nearby hard forks.The authors describe the relationship as a correlation and hypothesize the later gas-price uptick may reflect decreased Flashbots activity and a return to pre-Flashbots behavior.Source: Figure 6 · p. 8, §4.5 · p. 8
• Across months, more searchers submitted other transactions than any MEV type, by at least two orders of magnitude; searcher numbers for MEV types rose through August 2021, then declined and leveled out.The authors suggest many users left Flashbots for more profitable opportunities.Source: §4.5.1 · p. 8, Figure 7 · p. 9
• Other transactions were the most numerous, while sandwich and arbitrage counts tracked closely and liquidation transactions were far fewer.Transaction counts were more consistent than searcher counts, which the authors interpret as a distribution skewed toward a small number of searchers; they attribute fewer liquidations likely to fewer opportunities.Source: §4.5.1 · p. 8, Figure 7 · p. 9

5.1 Profit Distribution

• Flashbots’ benefit-distribution goal is to let searchers profit from MEV, with miners expected to take a lower share than non-miners.The measurements contradict that intended distribution; non-miners’ Flashbots income is more reliable but has lower expected value.Source: §5.1 · p. 9
• Miners averaged 0.125 ETH per Flashbots sandwich versus 0.048 ETH without Flashbots, while non-miners averaged 0.02 ETH versus 0.13 ETH without Flashbots.Profit standard deviations were 0.415 versus 0.127 for miners and 0.560 versus 0.154 for non-miners, respectively.Source: §5.1 · p. 9

5.2 Negative Profits

• 7,666 of 485,680 Flashbots-epoch transactions were unprofitable (1.58%), totaling 113.67 ETH or roughly 378,399.40 USD.The paper attributes these negative profits to faulty contracts submitted in searchers’ bundles, exposing searchers to losses.Source: §5.2 · p. 10

6 PRIVATE MEV EXTRACTION

• Taichi, Eden, and miner-created pools provide alternatives to Flashbots, but the authors could not identify how much MEV Taichi or Eden privately extracted.They therefore distinguish public, Flashbots, and non-Flashbots private extractions rather than attributing private MEV to individual pools.Source: §6 · p. 10

6.1 Discovering Private MEV Extractions

• The method labels included transactions absent from the publicly observed pending-transaction set as private, assuming the node observed most propagated transactions.It analyzes 774,725 blocks from November 23, 2021 to March 23, 2022; private MEV identification also uses public transactions as victims.Source: §6.1 · p. 10

6.2 Private MEV Distribution

• 81,089 of 99,928 sandwiches (81.15%) used Flashbots, while 18,839 (18.85%) occurred outside it.Figure 9 presents the distribution of private versus public MEV extraction.Source: §6.2 · p. 10, Figure 9 · p. 10
• Of the 18,839 non-Flashbots sandwiches, 13,238 (70.27%) were private; only 5,601 were public, or 5.6% of all 99,928 sandwiches.The authors conclude that Flashbots dominates private extraction, while other private pools coexist.Source: §6.2 · pp. 10–11

6.3 Private Miner MEV Extraction

• Two account addresses had private non-Flashbots sandwiches mined only by one miner each, suggesting two miners were extracting privately on their own.The paper identifies those miners as Flexpool and F2Pool participants and reports that both also mined private transactions associated with other account addresses.Source: §6.3 · p. 11

7 RELATED WORK

• Eskandari et al. proposed a frontrunning taxonomy, while projects including Flashbots and Eden Network formed private pools that let users bypass the public mempool.Daian et al. were first to introduce the term MEV and study frontrunning economically.Source: §7 · p. 11
• Daian et al. introduced MEV and studied priority gas auctions, while Torres et al. quantified historical frontrunning and Qin et al. [51] quantified MEV extraction.This paper leverages and extends their techniques to measure extraction using private pools such as Flashbots.Source: §7 · p. 11
• Piet et al. [50] and Capponi et al. [31] found miners earn most Flashbots profit, but did not compare that distribution with the pre-Flashbots period.This paper reports that searchers earned more before Flashbots and examines the existence and distribution of private pools.Source: §7 · p. 11
• Application-layer countermeasures include commit-and-reveal, slippage protection, and DEX design; consensus-layer proposals include fair ordering and transaction privacy.The paper says these approaches introduce costs or protect against only one attack type, and none are broadly adopted for large public blockchains such as Ethereum.Source: §7 · p. 11

8 CONCLUDING DISCUSSION

• Flashbots is not a one-stop MEV tool, increases Ethereum MEV transactions while reducing gas prices, and has benefited miners more than searchers.Users still need knowhow to design MEV-extraction solutions.Source: §8 · p. 11

8.1 Mission Accomplished?

• Goal 1’s status remains unclear: the authors could not access Flashbots’ pending pool, and its dashboard methodology is insufficiently open for comparison.They conclude Flashbots is more open than other private pools but does not seem optimal for illuminating MEV behavior.Source: §8.1 · p. 12
• Democratization remains ambiguous because extraction requires smart-contract and MEV expertise, while benefit distribution clearly falls short as non-miners earn less and usage declines.The authors say Flashbots offers more access to MEV infrastructure than before, but not as much as it could.Source: §8.1 · p. 12

8.2 Effects on Ethereum

• Flashbots lowers gas prices for users, but sealed-bid auctions raise searchers’ tips and fees, favoring miners.The authors attribute lower gas prices to separate transaction pools and auctions, while searchers increase tips because they cannot see competitors’ bids.Source: §8.2 · p. 12

8.3 Are We Going in the Right Direction?

• The authors question facilitating MEV extraction as a goal; critics call Flashbots-like services “theft” and a tax on ecosystem users.Source: §8.3 · p. 12
• The authors say Flashbots’ auction system helps miners compensate for reduced block rewards after the London hard fork.Source: §8.3 · p. 12
• Random ordering leaves a 25% chance of sandwich MEV, while cryptographic fair-ordering approaches are limited to small, reliable private blockchains.Source: §8.3 · pp. 12–13

• The study uses publicly available data voluntarily distributed by protocol participants and analyzes aggregates rather than individuals.Source: §9 · p. 13