ClearTrace

DEX Execution Insights

October 1, 2026  ·  Ethereum mainnet

Lead finding

A shortfall that does not change with trade size is not a price. It is a parameter.

We compare each venue's fill against the same trade routed directly through the Uniswap default, on the same pair, the same size and the same 30-minute window. Price impact is the thing that comparison is supposed to pick up, and price impact has a shape: it grows with size. A number that holds still from a $1,000 trade to a $1,000,000 one is therefore not reporting a price at all. It is reporting a setting somewhere in the route.

Three of the six venues we can currently compare on Ethereum show that shape. Two of the three are fully explained by a fee the venue publishes. The third is 1inch, and we still cannot account for it.

Median basis points of realized output against the Uniswap default route, Ethereum, matched on pair, size cohort and 30-minute window. Negative means the taker ends up with less than the default route delivers. 1,118 to 3,130 matched windows per cell, every cell rated: at least 30 realized samples spanning at least 7 days. Quote sample captured 2026-06-23 to 2026-10-01.

ParaSwap charges a flat 1 basis point on every major pair, paid to a named collector address, which we reproduced from our own sample on 2026-07-31. Its row reads -1.00, -0.90, -1.03 and -1.09, and its quote gap is exactly +1.00 bps at all four sizes. The fee is the whole of it. LI.FI charges a flat 25 basis points of the sell amount, a "LIFI Fixed Fee" taken before routing under its default API integrator, which we confirmed from a live quote on 2026-09-03 and published that week. Because that fee is taken from the sell token it sits inside both the quote and the fill, so LI.FI's quote gap stays near zero and the whole 25 bps surfaces here instead. Net of the published fee its routing lands at parity.

The third row we cannot account for, and it is pinned per pair rather than per venue. On USDC/USDT 1inch reads -9.93, -9.98, -9.99 and -9.95 basis points at $1k, $10k, $100k and $1M, on 354 to 374 matched windows per size, and from $10k upward the middle half of those windows sits within 0.2 bps of -10.00. On WETH/USDC it reads -28.70, -28.44, -24.62 and -9.38. Its own quote gap is 0.00 bps at every size, so 1inch delivers precisely what it quotes. The shortfall is in the quoted price.

This is not a new number and it is not a new reading of it. Our 2026-08-06 issue published the same cell under the same name, 28.97 bps behind the reference on WETH/USDC at $1k across 170 matched half-hours, and said then that our sampler could not attribute how much of a shortfall was a fee rather than a wider quote. Eight weeks of further sampling has moved the figure by a quarter of a basis point and has not answered that question. What we can rule out is an output-token fee: our fill decomposer measured a skim on 5,467 of 1inch's Ethereum rows and read a median of 0.0000 bps, so nothing is being taken out of what arrives. What we cannot rule out is either of the two remaining candidates, and they are not the same thing. A fee taken from the sell token before routing would be invisible in exactly this way, which is how LI.FI's hid from us for three months. So would a route built on a higher-fee pool for that pair, because a pool fee tier is also a fixed number of basis points and also does not move with size. The test above tells you the number is a parameter. It does not tell you whose.

The reason we cannot separate them today sits in our instrument rather than in the venue. Our collector asks the 1inch API for a quote and a gas estimate and does not ask it for the route breakdown it will return on request, so we hold 9,938 Ethereum captures and not one record of which pools any of them used. We also tried to pull a live quote while writing this and were refused from the analysis host, so that check has to run from the capture environment. That is the next thing we will do, and we will publish what it says whichever way it comes out. Until then the honest statement is the narrow one: the number is a parameter, we have not established whose, and we are not calling it a fee.

Also this period

The aggregator leaderboard, the quote-accuracy sample and the per-router revert breakdown are on the ClearTrace dashboard. The methodology notes set out the difference between the quote gap and the realized comparison, what the user-only revert rate filters out, and why revert rates are comparable only across router venues.

ClearTrace · Cross-chain DEX attribution. Quote and fill figures come from the forward-captured quote sample, 157,436 captures between 2026-06-23 and 2026-10-01, simulated on a fork that excludes gas, mempool competition, inclusion delay and real MEV, so they are a comparison of routes and not a full cost of trading. The realized comparison matches on pair, size cohort, chain and 30-minute window; the quote gap matches on 5-minute windows. Both are medians, and a median is not a worst case. Rows whose execution mode makes them unrepresentative are excluded from the comparison and counted, never silently dropped. Revert figures are on-chain over the 7 days to 2026-09-29 and are the user-only rate, which removes senders whose own transactions to these routers failed more than 10% of the time and is therefore bounded under 10% by construction. Attribution, coverage and volume figures are deliberately absent from this issue: that table last refreshed on 2026-09-21 and this week's story is about the quote sample. Every figure above is reproducible from ClearTrace's published data; the live leaderboard moves with each sync.