The Ledger
Ethereum PropAMMs Take Arbitrage Flow From Lagging AMMs
An Ethereum Research analysis of 92,627 fills across nine PropAMMs found that takers often traded against recent price moves while Uniswap lagged. The study says PropAMMs earned a small spread on trades that also used older AMM prices, but flags limits in its data and contract coverage.

Fresh quotes meet an older AMM price
Ethereum’s proprietary automated market makers are drawing trades that exploit price gaps with conventional pools: a new analysis of 92,627 fills found that PropAMMs tracked centralized-exchange moves within the block while Uniswap often lagged. The trades can give arbitrageurs a fresh venue for one leg and a slower-priced AMM for the other, while the PropAMM earns a small spread.
The analysis, posted on Ethereum Research on October 31, covers nine Ethereum PropAMMs over 30 days, from September 10 to October 9, 2026. It compares 69,827 WETH/stablecoin fills with Binance one-second trades. Of fills following a centralized-exchange move greater than 0.5 basis points in the preceding 12 seconds, 78.5% traded against that move—buying after a fall or selling after a rise.
That pattern does not mean takers made money on the PropAMM leg. The study reports an average taker-side markout of -0.45 basis points at the block timestamp and -0.74 basis points five minutes later. In the author’s measure, a negative markout means the subsequent reference price did not move in the taker’s favor.
Same-block updates are a contract rule
Three venues using the PrioUpdateRegistry—El Zorro, bopAMM and Kipseli—require the quote state to carry the current block’s timestamp. The analysis says their fills therefore execute against a maker update in the same block. Tempest accepts a wider timestamp window, while the study could not establish that FermiSwap’s same-block updates are enforced by its contract.
“Same-block pairing is therefore a contract rule, not evidence that makers are fast,” DanGuo, author of the Ethereum Research post, writes. The distinction matters: an on-chain timestamp requirement confirms which update a fill reads, but does not reveal when the maker created or signed that update off-chain.
The study’s broader dataset draws on Ethereum execution records from Xatu, contract identities from Sourcify and Binance spot trades as its price reference. Its TEMBRI commands can reproduce daily tables from public data, and the post says input files are pinned by hash. The author notes that Binance’s archived price is the last trade rather than a quote midpoint.
Arbitrageurs pair PropAMMs with slower pools
Transaction patterns suggest the fresh PropAMM quote often serves as one side of an atomic arbitrage. In a one-week review, 89% of direct-contract fills also swapped on Uniswap, Curve or Balancer. The post reports that 74% of those direct-contract transactions ended with no token lost and a gain, while the PropAMM collected its spread on the fill.
Among fills after a substantial exchange-price move, the gap between Uniswap and Binance widened with the move, while the PropAMM’s average markout stayed close to -0.4 to -0.5 basis points across the analyzed move bands. The author’s interpretation is that PropAMM pricing follows the centralized-exchange move in the same block, while the conventional pool’s price takes longer to adjust.
That comparison depends on the reference chosen. Measured against Uniswap’s last price instead of Binance, the average markout appeared positive for both buyers and sellers. The post cautions that Uniswap’s 0.05% fee band and routers’ choice to send trades to a PropAMM only when it beats the pool bias that comparison in the taker’s favor.
The sample leaves open questions
The study found no fills sandwiched on a PropAMM pool during the 30-day window. But the Flashbots public-mempool archive saw only 2.9% of PropAMM fill transactions; the author notes that “not seen” means absent from that archive, not proof that a transaction was private everywhere.
Coverage also varies by venue. FermiSwap, the largest venue in the sample, is decoded from BuilderNet’s description of its engine rather than verified source code; Tempest and Park are also not verified. About 4% of the observed swap logs came from unidentified contracts, and the test linking trades to arbitrage covers one week and excludes native ETH and payments to builders.
The author identifies two follow-up questions: why Metric fills can use oracle state with a median age of 132 seconds without visible adverse selection, and how much of the losses on lagging AMMs reaches PropAMMs as spread. The available results establish a measured pattern in the sampled trades, not a full accounting of arbitrage profits across Ethereum.
Sources
- Original source Ethereum Research ethresear.ch