The Invisible Leg: $132 Billion in Non-Atomic Arbitrage and the Eleven Searchers Behind It
Every twelve seconds, Ethereum’s DEX prices sit still while the rest of the world moves. Binance ticks. Coinbase updates. OTC desks reprice. By the time the next block arrives, the on-chain quote is stale - and someone is ready to close the gap.
That someone buys the underpriced asset on the DEX, sells it on the CEX, and pockets the difference. One leg on-chain, one leg off. Two separate systems, no atomic guarantee, real inventory risk. This is non-atomic arbitrage - and according to the first comprehensive study of its kind, it accounts for more than a quarter of all volume on Ethereum’s five largest DEXes.
The paper by Lioba Heimbach (ETH Zurich), Vabuk Pahari (MPI-SWS), and Eric Schertenleib identifies $132 billion in non-atomic arbitrage volume between the Merge and October 2023. It names the searchers doing it, quantifies their concentration, and uncovers a structural link between non-atomic arbitrage and the centralization of block building.
The Distinction That Matters
MEV research has historically focused on what happens within a single transaction or a single chain. Sandwich attacks, cyclic arbitrage, liquidations - all of these are atomic: every leg executes in one transaction, and if any leg fails, the whole thing reverts. No risk.
Non-atomic arbitrage is fundamentally different. The arbitrageur executes two trades on two separate systems:
- On-chain: buy the underpriced asset on a DEX (e.g., buy ETH with USDT on Uniswap)
- Off-chain: sell the same asset on a CEX (e.g., sell ETH for USDT on Binance)
These legs can’t be bundled atomically. The on-chain trade settles when the block is included. The off-chain trade settles on the CEX’s timeline. If the CEX price moves between commitment and execution, the arbitrage can become unprofitable. The searcher absorbs this inventory risk.
This is also why non-atomic arbitrage has been understudied: the off-chain leg is invisible. You can see the DEX swap on-chain, but the corresponding CEX trade leaves no trace on Ethereum. Identifying non-atomic arbitrage requires inference - matching on-chain behavior against off-chain price movements.
How It Works in the Block Building Pipeline
The flow follows the PBS pipeline with one critical addition - the off-chain leg happening in parallel:
-
Price divergence: between blocks, CEX prices move while the DEX quote stays frozen. The longer the gap, the larger the potential arbitrage.
-
Searcher constructs a transaction: recognizing the price difference, the searcher builds a swap to buy the underpriced asset on the DEX.
-
Submission to a builder: the searcher sends the transaction privately to a builder, often with a priority fee or direct payment.
-
Block construction and relay auction: the builder includes the transaction - typically at the top of the block, before any other swaps that might move the price - and bids in the relay auction.
-
The invisible leg: simultaneously or shortly after, the searcher executes the opposite trade on the CEX.
The first transaction in a block can close the entire price gap that accumulated during the previous twelve-second slot. This is why top-of-block position is the most valuable real estate in Ethereum - it’s where CEX-DEX price convergence actually happens.
The Numbers
Over the thirteen months between the Merge (September 15, 2022) and October 31, 2023:
| Metric | Value |
|---|---|
| Total non-atomic arbitrage volume | $132 billion |
| Share of top-5 DEX volume | >25% |
| Average daily value extracted | 112 ETH |
| Average trade size | ~$73,900 |
| Daily swaps identified | ~5,200 |
For context, the total volume on Ethereum’s five largest DEXes (Uniswap V2, Uniswap V3, Sushiswap, Curve, Balancer) over the same period was approximately $460 billion. Non-atomic arbitrage accounts for nearly a quarter of it.
Comparison with other MEV types
| MEV type | Average daily value (ETH) |
|---|---|
| Sandwich attacks | 161 |
| Non-atomic arbitrage | 112 |
| Cyclic arbitrage | 26 |
| Liquidations | 5 |
Non-atomic arbitrage is the second-largest MEV category by daily extracted value - comparable to sandwich attacks and far exceeding the on-chain-only categories that receive most of the research attention.
The Eleven Searchers
More than 90% of all identified non-atomic arbitrage volume flows through just eleven searcher addresses. The paper maps these to entities using bytecode analysis, operational timing patterns, and prior identification work.
The landscape is dominated by two clusters:
beaversearcher1 and beaversearcher2 - the largest non-atomic arbitrageurs by volume. Together they handle more swap volume than any other searcher group. beaversearcher1 is the only major searcher operating continuously throughout the entire thirteen-month data collection period. beaversearcher2 came online later, during a period of higher market activity.
rsyncsearcher1, rsyncsearcher2, and rsyncsearcher3 - the second-largest cluster. These searchers rotated through operation: rsyncsearcher1 ran first, then rsyncsearcher2 took over, then rsyncsearcher3. At any given time, typically one was active.
The remaining identified searchers include jumpersearcher, mantasearcher, builder1searcher, and several others labeled searcher1 through searcher3 in the paper. Each is a distinct on-chain address performing non-atomic arbitrage with consistent behavioral patterns.
What they trade
The searchers concentrate on pools between major token pairs: ETH, BTC, USDC, USDT, and DAI. This makes sense - non-atomic arbitrage requires a liquid off-chain market to execute the second leg, which limits the strategy to tokens with deep CEX liquidity. The paper finds that larger searchers have larger median trade sizes, with some averaging over $100,000 per swap.
How the market share evolved
The paper tracks market share over time and finds notable patterns:
- beaversearcher1 accounts for at least 10% of total volume in every period, but their share spikes during high-volatility episodes (with peaks coinciding with cryptocurrency price volatility)
- The rsyncbuilder searcher cluster shows market share that varies with which specific searcher address is active
- Several searchers appear to enter the market during high-volatility periods and reduce activity during calmer markets
On most days, the two largest searchers alone account for more than half the non-atomic arbitrage volume. The market is not just concentrated - it’s dominated by a handful of operators with the infrastructure to simultaneously execute trades across on-chain and off-chain venues at speed.
The Vertical Integration Finding
This is the paper’s most structurally significant discovery: four of the eleven searchers are operationally integrated with block builders.
| Builder | Integrated searcher(s) |
|---|---|
| beaverbuild | beaversearcher1, beaversearcher2 |
| rsyncbuilder | rsyncsearcher1, rsyncsearcher2, rsyncsearcher3 |
| builder1 | builder1searcher |
| mantabuilder | mantasearcher |
The evidence for integration comes from multiple signals: shared bytecode between searcher and builder contracts, correlated operational timing (the searcher and builder start and stop operating at the same time), and statistically significant correlation between the searcher’s non-atomic arb volume share and the builder’s block market share.
Why integration matters
When the builder is also the searcher, the economics change fundamentally:
The builder doesn’t need to share MEV revenue with an external searcher. In a standard PBS flow, the searcher captures the arbitrage profit, pays a portion to the builder as a priority fee, and the builder passes most of it to the proposer through the relay auction bid. With vertical integration, the builder captures the full value chain - the on-chain priority fee, the off-chain CEX profit, and the strategic advantage of guaranteed inclusion.
The builder can bid higher in the relay auction. Because the integrated builder’s effective revenue includes off-chain CEX profits that don’t appear on-chain, they can afford to bid more aggressively. External builders who only see on-chain fees can’t compete during periods when non-atomic arb is highly profitable.
The correlation is measurable. The paper finds statistically significant correlation between the share of non-atomic arbitrage volume handled by integrated searchers and the share of blocks won by their corresponding builders. For beaverbuild, the correlation has a p-value of 1.83 x 10^-6. For rsyncbuilder during the operation of rsyncsearcher2, the correlation is 0.805 with a p-value of 5.68 x 10^-18.
The Beaverbuild Subsidy
The paper’s most striking data point concerns beaverbuild’s on-chain profitability.
Between February 3 and March 14, 2023, beaverbuild’s daily builder profit - priority fees and coinbase transfers received minus the payment to the proposer - was consistently negative. The builder was paying proposers more than what transactions in its blocks paid in fees. Over the full study period, beaverbuild accumulated an on-chain loss of 3,195.76 ETH.
During this time, beaverbuild built 48,606 blocks. Of these, 73.2% offered the proposer a higher value than the builder received in transaction fees. And 97.46% of those subsidized blocks contained at least one transaction from beaversearcher1 or beaversearcher2.
The implication is straightforward: beaverbuild was subsidizing block building with revenue from its integrated searchers’ off-chain CEX trades. The on-chain loss is a lower bound on the off-chain profit - the builder could afford to lose 3,195 ETH on-chain because it was making more than that on CEX legs that never touch the blockchain.
This finding has a broader significance beyond one builder’s strategy. It demonstrates that on-chain builder profitability data is structurally misleading for integrated searcher-builders. Any analysis of builder economics that only examines on-chain flows misses the off-chain revenue that funds the operation.
Volatility Is the Revenue Driver
Non-atomic arbitrage volume is strongly correlated with cryptocurrency price volatility:
| Correlation | Value | Significance |
|---|---|---|
| Non-atomic arb volume vs. ETH daily volatility | 0.725 | p < 10^-10 |
| Non-atomic arb volume vs. BTC daily volatility | 0.729 | p < 10^-10 |
The relationship is intuitive: more volatility means larger price divergences between blocks, which means larger arbitrage opportunities. But the data reveals several additional patterns:
Volume concentrates during US market hours. Non-atomic arb activity peaks between 14:00 and 21:00 UTC (Monday through Friday), corresponding to US equity market hours and peak CEX liquidity. A secondary peak appears around 08:00-10:00 UTC during the Asian-European overlap.
Extreme events produce spikes. The FTX collapse (November 2022), the USDC depeg (March 2023), and Federal Open Market Committee announcements all produced visible surges in non-atomic arb volume. These aren’t gradual increases - they’re sharp spikes where relative price changes exceed 0.66% and 1.3% for ETH and BTC respectively in under ten seconds.
HFT builders win more blocks during high volatility. The correlation between BTC/ETH price volatility and the share of blocks won by HFT builders (those with integrated searchers) is 0.783 for ETH volatility and 0.650 for BTC volatility. During the periods that matter most economically, integrated builders dominate block production.
The Block Space Impact
Non-atomic arbitrage’s footprint on block space is modest most of the time but significant during the moments that matter most.
Gas usage: in more than 99% of blocks, non-atomic arb uses less than 5% of available gas. But during the 99.9th percentile of BTC price volatility, approximately 40% of blocks have non-atomic arb consuming more than 10% of gas.
Block value capture: during those same high-volatility periods, non-atomic arb transactions account for more than 50% of total block value (priority fees plus direct builder payments) in roughly half of all blocks. In some blocks, the share exceeds 80%.
This concentration matters because high-volatility periods are precisely when block space is most scarce. Regular users face higher base fees and longer inclusion delays during the same windows when non-atomic arb transactions are consuming the most gas and block value. The system-level cost falls on users who have no connection to CEX-DEX price dynamics.
How They Identified Non-Atomic Arbitrage
Since the off-chain leg is invisible, the paper develops five heuristics to identify the on-chain leg. A transaction must satisfy all five to be classified as non-atomic arbitrage:
- Simple swap: executes a single swap on one DEX - no multi-hop routing or bundled operations
- Private transaction: not broadcast to the public mempool - submitted directly to a builder
- First swap in pool: the first swap in the relevant pool in that block - claiming the top-of-block price correction
- Priority payment: includes a coinbase transfer to the builder or has a priority fee of at least 1 Gwei
- Established tokens: involves two tokens with reliable off-chain pricing (traded on CEXes)
The authors validate these heuristics against the known searcher addresses. For all major searchers except one, more than 80% of their transactions satisfy all five criteria. The one exception (searcher2) appears to mix non-atomic arbitrage with simpler swaps, producing a lower match rate.
The methodology likely undercounts total non-atomic arbitrage. Failed attempts, transactions that don’t meet all five criteria but are economically motivated by CEX-DEX price differences, and non-atomic arb on DEXes outside the top five are all excluded.
Connection to LVR and the MEV Pipeline
Non-atomic arbitrage is the market mechanism through which Loss-Versus-Rebalancing (LVR) is realized on-chain. LPs on AMMs lose value because their quotes go stale between blocks. Non-atomic arbitrageurs are the ones who exploit that staleness - buying cheap on the DEX, selling at the real price on the CEX.
This creates a direct economic chain:
- LPs lose through stale quotes (LVR)
- Non-atomic arbitrageurs capture that value through CEX-DEX trades
- Builders receive priority fees from the searchers’ on-chain leg
- Proposers receive the builder’s bid through the relay auction
The LVR paper proved that constant block times minimize LP losses for a given average block time, and that halving block time reduces time-normalized LVR by approximately 29%. This paper provides the downstream measurement: 112 ETH per day extracted from LPs through the non-atomic arb mechanism, flowing through the PBS pipeline to proposers.
If sub-slots or shorter block times reduce LVR (as the theory predicts), non-atomic arb revenue per event should decrease proportionally. The CEX-DEX research presented at the Blockspace Forum Workshop aligns with this: shorter execution windows reduce per-trade profits but increase trading volume, producing a net reduction in LP losses with diminishing returns at shorter intervals.
Proposed Mitigations
The paper discusses two directions:
Separating top-of-block
Non-atomic arbitrage concentrates at position zero in the block. If the top-of-block slots were auctioned separately from the rest of the block body, integrated builders would lose their leverage over full block construction. They could still compete for the top-of-block arbitrage slot, but the commodity transaction portion of the block could be built by a different party.
This aligns with ideas like PEPC-Boost (top-of-block and rest-of-block bidding) and, more broadly, with the relay block merging mechanism - which already creates a separation between the winning builder’s block (including top-of-block MEV) and the non-contentious transactions appended at the bottom.
Shorter block times
Reducing the interval between blocks shrinks the window for CEX-DEX price divergence, making each arbitrage opportunity smaller. But even with one-second blocks, the first transaction still captures whatever drift occurred. The top-of-block problem remains at any block time.
For Block Builders
Non-atomic arbitrage is the single largest source of top-of-block value. At 112 ETH per day and more than a quarter of top-DEX volume, it dwarfs cyclic arbitrage and liquidations. Understanding which pools generate the most non-atomic arb flow - and when - is fundamental to block value optimization.
Vertical integration between searching and building is a structural advantage, not a tactical one. The paper demonstrates that integrated builders systematically outbid non-integrated builders during high-volatility periods. The off-chain CEX profit subsidizes on-chain bidding, creating a competitive dynamic that can’t be replicated by builders who only see on-chain transaction fees.
On-chain profitability metrics are misleading for integrated operations. Beaverbuild’s 3,195 ETH on-chain loss wasn’t a failed business - it was a subsidy funded by off-chain revenue. Any competitive analysis of builder economics that relies solely on on-chain data is seeing less than half the picture.
Volatility is the variable that matters most. Non-atomic arb revenue is convex in price changes - calm markets produce modest returns, volatile markets produce outsized ones. Builder revenue forecasting needs to account for volatility regimes, not just average daily extraction rates.
The top-of-block / rest-of-block split is already happening. Relay block merging operates on non-contentious bottom-of-block transactions. Non-atomic arb operates at the top. These are economically distinct regions of the block with different competitive dynamics. Builders who understand which of their flow is top-of-block sensitive and which is commodity-like can optimize for both.
Sources
Original paper:
- Non-Atomic Arbitrage in Decentralized Finance - Lioba Heimbach (ETH Zurich), Vabuk Pahari (MPI-SWS), Eric Schertenleib. January 2024.
Referenced prior work:
- Quantifying Blockchain Extractable Value: How dark is the forest? - Gupta et al. The searcher identification methodology this paper builds on.
Related Rawcall articles:
- Why Fixed Block Times Are the Best Deal LPs Already Have - the LVR framework that quantifies LP losses upstream of non-atomic arbitrage
- What Emerged from the Blockspace Forum Workshop - sub-slot execution and CEX-DEX research
- Relay Block Merging: How Losing Builders Still Get Paid - the mechanism that separates top-of-block from bottom-of-block economics
- Ethereum’s Blockspace Market Under the Microscope - the exclusive order flow dynamics that non-atomic arb amplifies