On a new token, the question is not the price. It is whether the trade goes through.

Every execution comparison we publish, ours included, starts from a quote that became a fill. We ask a venue for a route, simulate that exact route on a fork of mainnet, and measure what the taker receives. On the four Ethereum majors the step from quote to fill is close to free: across 46,992 captures on WETH/USDC, WETH/USDT, WBTC/USDC and USDC/USDT, 43,422 of them realized, 92.4%, and only 2.01% were routes that reverted when we simulated them. That is the regime where a one basis point difference between venues is worth arguing about.

Off the majors it is a different market. Across the ten other Ethereum pairs we sample, 28,568 captures, 78.2% realized and 5.8% were routes that reverted. The worst pair in the sample is FWA/USDC, and it is worth reading in full. We captured 2,534 quotes for it between August 11 and September 17, across nine venues, at $1k and $10k. 419 of them became a fill. That is 16.5%. Of the misses, 1,184 were quotes we never obtained at all, where the venue returned an error or no route, and 907 were routes we did receive and that reverted in simulation.

The failures are not spread evenly. Seven of the nine venues produced no fillable route for this pair at any point in five weeks, each on 282 captures except one, which sampled 278. Two produced fills, and every one of the 419 fills, without exception, was routed through a Uniswap v4 pool. Even the direct Uniswap route, which reaches those pools by construction, filled 157 of the 282 routes we simulated for it. The liquidity exists and is reachable. Most of the routing layer was not reaching it.

Ethereum pairs	Captures	Became a fill	Reverted in sim
The four majors	46,992	92.4%	2.01%
The other ten pairs	28,568	78.2%	5.8%
V4/USDC	2,790	54.6%	7.0%
FWA/USDC	2,534	16.5%	35.8%

Ethereum, all size cohorts, quotes captured 2026-06-23 to 2026-09-17. A capture becomes a fill when the venue returns a route and that route completes in a fork simulation. The remainder splits into quotes never obtained, off-chain RFQ legs we cannot simulate, venues our fork cannot reproduce, our own build failures, and routes that reverted. Only the last of those is counted in the right-hand column.

Two honest limits on this. A route that reverts in our simulation is not proof that the same trade would revert on chain: the fork has no competing traders in it, and a stale quote fails in a simulator for reasons a real mempool would resolve differently. And a quote we never obtained is at least partly our own problem: an API that rate limits us reads the same as one that has no route. What survives both caveats is the shape. On majors the routing layer converges and the argument is about basis points. On a five week old pool it diverges completely, and the venue you happened to ask decided whether you traded at all.

Also this period

At $1M, most venues land within 3 bps of simply using the Uniswap default. Matched on pair, size and time window, five of the venues shown here fill inside that band: KyberSwap +2.57 bps, Sushi +1.32, OKX +0.07, OpenOcean -0.96, ParaSwap -1.05, over 698 to 1,229 matched windows each. OpenOcean and ParaSwap both charge a fee taken from the output token, 1 bp on stables and 10 bps on volatile pairs for the first and a flat 1 bp for the second, and those fees are already inside the figures above. Of the two named venues outside the band, LI.FI reads -20.40 bps because LI.FI charges a flat 25 bps fee taken from the sell token before routing, which puts the fee in the fill rather than in the quote. The other is the venue we left unnamed on September 3, pending its own right of reply, and it still is. Not every sampled venue at this size is shown: one more clears the bar for a rated cell but is withheld here, for reasons unrelated to its numbers.

84,968 sandwich attacks in the 7 days to September 16, against $73.3M of victim volume across the four chains we track. The count and the value point at different chains: Base carried 75,032 of the attacks for $16.0M, Ethereum 9,439 for $57.0M. Among the 500 most recent victim trades in the sample the median size is $133.58, so the Base figure is a very large number of very small trades.

User-submitted revert rates on Ethereum routers stayed at or under 1.3% over the same 7 days. 1inch 0.18% on 117,049 user transactions, LI.FI 0.32% on 32,316, Sushi 0.38% on 37,715, ParaSwap 0.39% on 29,365, KyberSwap 0.81% on 185,532, Uniswap 1.02% on 123,330, OKX 1.3% on 28,863. These strip senders classified as bots and searchers, which is most of the distance between a headline rate and this one: OKX's headline reads 6.10% against 1.3% here, Uniswap's 2.53% against 1.02%, KyberSwap's 1.52% against 0.81%. The comparison holds only across router venues, where the user submits the swap, so a batch auction or an intent venue does not belong in the list. Three further routers are measured and are not named here, each held back by a check about our own evidence rather than by anything in its numbers: one is withheld pending its own right of reply, and for two we cannot yet meet our own bar for naming a venue. All three sit inside the range above, so their absence does not move it.

The quote sample now stands at 131,436 captures over 86.5 days, 112,023 of them realized in a fork simulation, 85%. 344 of 426 venue, chain and size cells are rated, meaning at least 30 realized samples spanning at least 7 days. One venue's cell is marked stale rather than rated: we have been unable to collect OpenOcean quotes since August 21, which is our access and not that venue's availability.

The aggregator leaderboard, the per-router revert breakdown and the quote-accuracy sample are on the ClearTrace dashboard. The methodology notes set out how the matched quote gap, the matched realized comparison and the rated-cell bar are computed, and why the two comparisons are never interchangeable.

Open Dashboard

Methodology

ClearTrace · Cross-chain DEX attribution. Quote figures come from the forward-captured quote sample, 131,436 captures between 2026-06-23 and 2026-09-17, matched against the Uniswap default route by pair, size cohort, time window and chain, on 30-minute windows for the realized comparison. That comparison excludes gas, mempool competition, inclusion delay and real MEV, and cannot reproduce off-chain RFQ legs, so it is not a complete all-in trading cost. Where a venue charges an explicit fee, that fee lands inside one of the two comparisons and not the other, depending on whether it is taken from the sell token or the output, so neither number is by itself evidence about routing quality. Rows whose fill came from a non-default execution mode are excluded from the realized comparison and the exclusion is reported with the result. Three sampled venues are absent from the $1M comparison above: one whose quotes are too often priced in a non-standard mode to be set against standard ones, one that has wound down its API, and one withheld for reasons unrelated to its execution numbers. Sandwich and revert figures are on-chain over a 7-day window ending 2026-09-16. The user-only revert rate removes senders classified as bots or searchers, and is comparable only across router venues; a batch auction or an intent venue clears off-chain, so an unfilled order never reverts on-chain. Attribution and coverage figures are deliberately absent from this issue: that table last refreshed on 2026-09-14 and we do not publish a moving number off a frozen one. Every figure above is reproducible from ClearTrace's published data; the live leaderboard moves with each sync.

Get these findings in your inbox

Periodic data on DEX execution quality, aggregator benchmarks, and attribution coverage. No spam.