NewsCryptoHow Proprietary AMMs Execute Trades on Ethereum and Manage Fill Risk

How Proprietary AMMs Execute Trades on Ethereum and Manage Fill Risk

Author: DefiLiban·

Key Takeaways

  • •PropAMMs on Ethereum have accumulated over $3.35 billion in lifetime volume across 277,834 swaps, with Fermi and Metric together trading about $460 million during the week of September 14–20, 2026, approximately 7.6% of Uniswap's $6.0 billion weekly volume.
  • •Market makers face a blind window before learning of fills, with median latency of 1.23 seconds, and in 16% of measured trades the Binance reference price moved beyond the quoted 0.90 bps half-spread during that period.
  • •The guarantee that a maker's latest price update precedes a taker's swap relies on coordination among four builders—Titan Builder, Quasar, BuilderNet and Bombora—rather than Ethereum protocol rules, and the proposed ERC-8324 standard is not yet live.
  • •Titan Builder produced 44% of all Ethereum blocks during the measured period while carrying 69% of PropAMM fills, highlighting concentration in the model's execution infrastructure.
  • •Measured ETH/USDC full spreads ranged from 0.86 basis points at Metric to 4.20 basis points at bopAMM, with Fermi at 2.34 basis points, reflecting differing maker calibrations of adverse-selection risk.
How Proprietary AMMs Execute Trades on Ethereum and Manage Fill Risk

Proprietary automated market makers (PropAMMs) on Ethereum have processed more than $3.35 billion in all-time volume across 277,834 swaps, according to independent researcher azfuller. Two venues alone traded nearly $460 million during one week in September 2026 and reached a peak 11% share of Ethereum decentralized-exchange volume.

A post-mortem published on ethresear.ch on November 19, 2026, by santidipaolo, jpinasco and agustinpolfinance of POL Finance examines how these fills occur and the unhedged exposure market makers face between signing a quote and receiving the block containing the trade.

What a PropAMM Is

A PropAMM is an on-chain liquidity pool whose quoted prices are continuously overwritten by a single market maker, potentially hundreds of times per block. Unlike a constant-product pool based on the x*y=k formula, where reserves are passive and any taker can trade against the curve, a PropAMM operator sets executable prices using inventory, real-time market data and risk limits. The quotes are then published to supported block builders.

For traders, the key distinction is that a PropAMM fill is based on a firm quote rather than a reserve ratio. The maker determines the executable price at any given moment, can widen spreads when inventory becomes unbalanced, and can withdraw quotes altogether. Instead of leaving capital across the full Uniswap v3 range, the model concentrates liquidity where demand exists.

Azfuller’s data memo discusses capital turnover at Fermi and comparable Uniswap v3 pools, but does not specify the observation date, measurement window, pool-selection method, liquidity definition or turnover calculation used for that comparison.

According to LambdaClass’s protocol blog, Titan Builder published a PropAMM API in May 2026. The documented API description does not establish an Ethereum protocol or client version, contract deployment block, block timestamp or effective block for the API state. Quasar, BuilderNet and Bombora later adopted the same API.

Those four builders now carry all PropAMM venues. The model depends on an ordering guarantee under which the maker’s most recent price-update transaction is placed before the taker’s transaction in the same block. That guarantee is enforced through cooperation among the builders rather than by Ethereum protocol rules. As the ethresear.ch post states: “That ordering is a promise by four companies, enforced by nothing in the protocol.”

Key Measurements and Infrastructure

PropAMMs rewrite on-chain prices hundreds of times per block using a single maker’s inventory. The maker remains unaware of a fill for a median 1.23 seconds between signing a quote and seeing the executed block. During that window, 16% of measured fills saw the reference price move beyond the quoted half-spread.

Ordering guarantees currently depend on promises from the four builders. ERC-8324 has been proposed to formalize the mechanism at the protocol level, but it is not yet live.

Kubi Mensah of Gattaca, Titan Builder’s parent company, described the model’s structural advantage as follows: “The beautiful thing about prop AMMs [is] that they are on the ledger, and then anyone that interacts with the chain…can discover best prices,” according to azfuller’s memo.

LambdaClass identifies a PropAMMRouter contract at 0x4DdF368080CD7946db5b459aD591c350158175e1. The router covers all PropAMM venues, includes a Uniswap v3 fallback and charges zero fees. Its intent-routing design is similar in spirit to the infrastructure built by CoW Swap for MEV-protected intent trading.

From Quote to Ethereum Settlement

Quote formation

The maker operates an off-chain pricing engine that continuously calculates bid and ask quotes based on current inventory, a reference feed such as the mid-price of Binance perpetuals, and a target spread intended to cover adverse-selection risk during the blind window. The resulting quotes are signed off-chain and submitted to supported block builders. A newer signature can overwrite an earlier one before the block is sealed.

The POL Finance analysis measured ETH/USDC spreads across live venues. Metric quoted a full spread of 0.86 basis points, Fermi quoted 2.34 bps and bopAMM quoted 4.20 bps.

Order acceptance and risk checks

In a hypothetical USDC-to-ETH swap routed through the PropAMMRouter, the router queries each venue for a live quote, selects the best executable price and submits a transaction referencing the maker’s current signed price. The builder places the maker’s latest price-update transaction immediately before the taker’s swap in the block, locking in the quoted rate.

No separate on-chain oracle or pool-reserve calculation is required. The fill price is the price specified by the maker’s latest signed message. Partial fills, quote expiry and transaction reverts are all possible.

If the latest signed quote has expired or has been superseded by an update with a wider spread, the transaction reverts. The user can then fall back to the Uniswap v3 route included by the PropAMMRouter as a safety mechanism. Fees are reflected implicitly in the spread; PropAMMs do not use explicit fee tiers such as Uniswap v3’s 0.05%, 0.30% and 1.00% pools.

On-chain settlement and the blind window

Settlement occurs fully on-chain. Assets transfer atomically when the block is accepted, and any observer can verify the fill. The maker’s principal risk arises between signing the quote and seeing the block that contains the trade.

The POL Finance analysis measured that interval directly. The maker learns that a fill occurred only when the block propagates to its own node. Median latency was 1.23 seconds, while latency at the 90th percentile was at least 2.18 seconds. During this blind window, the maker cannot hedge the position assigned by the fill.

The researchers quantified the resulting exposure. In 16% of measured trades, the Binance reference price moved beyond the 0.90 bps half-spread during the blind window. In those cases, the fill occurred at a price that the market had already moved through. Over time, the spread must compensate for this per-trade adverse-selection rate.

Titan Builder produced 44% of all Ethereum blocks during the measured period but carried 69% of PropAMM fills. This concentration gives makers more predictable execution timing when Titan is the winning builder.

Trade-Offs for Traders and Market Makers

For traders, the main benefit of a PropAMM fill is price certainty within the quoted size. Because the maker concentrates capital near the current market price instead of distributing it across a complete curve, small- to medium-sized swaps can experience less price impact than they would in a comparable passive pool.

Azfuller’s data memo recorded $280 million in weekly volume for Fermi and $178 million for Metric during the week of September 14–20, 2026. Together, those figures represented approximately 7.6% of Uniswap’s $6.0 billion in weekly volume and indicated taker demand for this execution model during its early adoption.

The limitations are structural. Maker capacity is finite and constrained by inventory. A PropAMM can widen its quotes or go offline when the maker’s risk limits are breached. A withdrawn quote also means that a taker may see an attractive price in a simulation that is no longer available when the transaction executes.

This behavior differs from that of a passive pool, which remains executable at some price regardless of market conditions. PropAMMs also introduce a dependency on builder cooperation that passive AMMs do not have. If a builder fails to honor the ordering guarantee, the maker’s price update and the taker’s swap can appear in the wrong order, potentially producing a fill at a stale price.

According to the POL Finance ethresear.ch post, ERC-8324 has been proposed to enforce the ordering mechanism at the protocol level, but the standard is not yet live. Until it is implemented, PropAMM participants rely on a coordination agreement among four companies rather than on a smart-contract invariant.

Before routing significant volume through a PropAMM, traders and integrators should examine whether the venue publishes its quote-expiry and fallback logic, the maker’s stated inventory limits for each token pair, the builders carrying the venue and the share of recent blocks those builders produced, as well as the smart contract’s audit history and upgrade mechanisms.

The spread figures—0.86 bps at Metric compared with 4.20 bps at bopAMM—also show how different makers calibrate adverse-selection risk against capital deployment. Comparing spreads across venues over time provides an on-chain measure of competitive execution quality.