|
ClearTrace
DEX Execution Insights
August 6, 2026 · Ethereum mainnet
|
|
Lead finding
A perfect quote-accuracy score tells you nothing about the price you got
ClearTrace polls live quotes on Ethereum majors at four trade sizes, then fork-simulates the route each quote returns. The number that falls out is a venue's gap to its own quote: how far the fill drifted from what it promised. It measures whether a venue is honest about its own pricing, and it is routinely read as a measure of who is cheap. We have drawn that distinction in these pages before. This is the sharpest evidence we have had for it. Below is the gap on WETH/USDC, and then what the same orders actually paid out.
| Gap to own quote (bps) |
$1k |
$10k |
$100k |
$1M |
| Uniswap (venue reference) | 0.00 | 0.00 | 0.00 | 0.00 |
| Sushi | 0.00 | 0.00 | 0.00 | 0.00 |
| Bebop (self-execution mode, see note) | 0.00 | 0.00 | 0.00 | 0.00 |
| 1inch | 0.00 | 0.05 | 0.09 | 0.22 |
| ParaSwap | 1.00 | 1.00 | 1.00 | 0.99 |
| LiFi | 1.09 | 1.20 | 2.56 | 1.28 |
| KyberSwap | 0.00 | 3.72 | 4.11 | 3.36 |
| OpenOcean | 10.01 | 10.12 | 10.01 | 10.01 |
| OKX | 1.29 | 4.99 | 6.80 | 17.10 |
Rated cells only: at least 30 realized simulations each, 102 to 178 per cell, spans of 26 to 44 days. Uniswap is the direct-pool reference and sits at 0.00 by construction, since its quote and the simulation run the same pool math. Bitget is on-chain only, never quote-sampled. Odos wound down on 2026-07-30 and is held out; on the realized-output measurement below it was ahead of the pool at $1k, so holding it out flatters that comparison rather than the reverse. One further sampled venue is held out of published tables, and its figures would not change any finding here.
Read that table as a price ranking and you would take 1inch over LiFi at $1,000: a flat 0.00 against 1.09 bps. Now take the same orders, matched to the same half hour at the same notional, and count the tokens that came back against Uniswap's own quote for the same order. 1inch delivered 28.97 bps less across 170 matched half-hours. LiFi delivered 25.00 bps less across 108. Compared head to head on the 107 half-hours where both quoted, LiFi delivered 3.41 bps more than 1inch. The venue with the better quote-accuracy score handed the taker less money.
A quote can be scrupulously honest and still be struck on worse terms. A venue that takes 25 bps and then fills exactly where it said scores a perfect 0.00 on the first measurement while being 25 bps behind on the second. The sharpest version of this is the three venues that post an identical 0.00 at $1k. Sushi, KyberSwap and 1inch cannot be separated on quote accuracy at all, and their realized shortfalls against the reference are 0.00, 0.00 and 28.97 bps. So we publish the gap to a venue's own quote as an honesty measure, never as a cost ranking. We should add that our sampler cannot yet attribute how much of each shortfall is an explicit fee rather than a wider quote, so the 25 bps above is measured, and calling it a fee would not be.
|
|
|
This period in numbers
| |
Three of the seven get ahead of the reference at $1M
Same matched measurement on WETH/USDC, against Uniswap's own quote. At $1,000 every venue we publish is at or behind the pool: OpenOcean 10.26 bps behind and never ahead across 133 half-hours, LiFi 25.00 behind, 1inch 28.97 behind, with Sushi, KyberSwap and OKX level at 0.00. At $1M three pull ahead: KyberSwap 6.00 bps ahead in 77.8% of 144, OpenOcean 3.85 ahead in 65.6% of 128, Sushi 1.30 ahead in 91.5% of 130. The other four stay behind, LiFi still 20.54 behind. Two cautions on the $1M column, both of which cut against the finding: KyberSwap realizes only 85.8% of its $1M quotes and 97.4% of the fills that survive are RFQ-routed, a leg the simulation cannot reproduce, and 17.0% of 1inch's $1M samples end in a simulated revert and drop out, though 1inch still trails the pool even after that filtering. For KyberSwap, the venue whose worst cases drop out is the one that looks best; treat its $1M lead as softer than the others cited here. Source: ClearTrace quote sampler, Ethereum, rated cells.
|
| |
We found a 1% error this week, and it was ours
While writing this issue we found a roughly 1% error in our own Bebop series and traced it to our collector. On WETH/USDC, Bebop's realized output sat 95.3 to 99.1 bps below the Uniswap reference across the four sizes. Three of its other pairs run the same way, USDC/USDT and WETH/USDT near 95 to 100 bps and WBTC/USDC out to 133 at $100k. Two run smaller, SPX/USDC at 71 to 87 and PEPE/USDC at 63 to 67. One is close to parity, and it is the one that explains the rest: on ASTEROID/USDC, where 100% of Bebop's samples come back on its auction route, the shortfall is only 3.3 bps at $1k and 11.9 at $10k. Our sampler requests quotes in self-execution mode, which on most pairs returns Bebop's market-maker route rather than its auction route, and that route quotes materially wider. Re-requesting the same three orders live on 2026-08-06, the standard mode came back 100.08 bps better at $1k, 98.75 at $10k and 108.52 at $1M. Our own history points the same way: on the WETH/USDC samples where the auction route did come back, Bebop's median sat 0.12 bps from the Uniswap reference across 151 half-hours, which is parity. Two honest limits on that diagnosis. The live re-request is three quotes on one day. And the auction route sometimes appeared under the same self-execution request earlier in the summer and stopped, so the request mode is not the whole mechanism. What we can say is that the shortfall tracks which route comes back, so we are publishing no price comparison for Bebop until the collector is fixed. Its quote-accuracy row stands, since that compares Bebop only against its own quote. One separate limit, which bears on the reference itself: on the thin pair SYN/USDC, all ten sampled venues price hundreds to thousands of basis points above the Uniswap reference, so there the reference is the outlier and the venues agree with each other. We publish no SYN/USDC figures. It is a standing reason to read the reference as one venue's route, which is all it claims to be. Source: ClearTrace quote sampler, plus a live re-request on 2026-08-06.
|
| |
User revert rates, counted only where a failed swap is a real reverted transaction
Across the router-model venues, where the trader submits the swap so a failure is a transaction we can count, the genuine-user revert rate (methodology v6) ran from 1inch 0.26% and Sushi 0.30% up to Uniswap 1.43% among the live venues, with the retired Odos higher still at 2.52% over its final stretch of data. We do not rank revert across execution models: on a batch auction like CoW or an intent venue like 1inch Fusion a solver submits, so an unfilled order never becomes a transaction and its on-chain revert rate is structurally near zero rather than a reliability signal. Source: ClearTrace aggregator leaderboard, onchain_user_revert_rate_pct.
|
| |
What the read rests on
The sampler has taken 41,970 forward-captured quote simulations over 44.4 days across four chains, 92% of which reached a realized fill. This issue is a narrow slice of that: Ethereum WETH/USDC, about 6,000 captures and 5,799 realized fills. A cell is a chain-venue-pair-size combination, and we cite one only once it clears a rated bar of at least 30 realized simulations and a 7-day span. Counted per chain, 281 of the 660 cells we have sampled clear it today. Every figure in the table and in the first bullet comes from a rated cell; the two exceptions are flagged where they appear, being the live re-request in the Bebop note and that note's cross-size parity figure. Source: ClearTrace quote sampler collection log.
|
|
|
|
The full aggregator leaderboard, the per-pair comparison and live chain-by-chain coverage are on the ClearTrace dashboard. The methodology notes explain how the quote gap, the matched realized-output comparison, the user-only revert rate and the rated-cell bar are computed.
|
|
ClearTrace · Cross-chain DEX attribution · Quote-sampler and aggregator-leaderboard data, Ethereum, 2026-08-06 snapshot. The gap to own quote is a venue's quoted output against the output a fork simulation of its returned route produces. It measures a venue against its own promise and carries no information about whether that promise was competitively priced. The realized-output comparison pairs each venue's simulated fill with Uniswap's own simulated fill from the same half hour, at the same notional and direction, and reports the median across matched rounds. Uniswap is a reference point, not an unrouted baseline: its quote is served by Uniswap's own routing API, which split the order across v3 and v4 pools in 401 of 691 WETH/USDC samples. Neither measurement includes gas. A fork simulation excludes gas cost, mempool competition, inclusion delay and real MEV, and it cannot reproduce off-chain RFQ legs, so none of these figures is a complete all-in trading cost. Rated means at least 30 realized samples and a 7-day span. Every cell cited in the table and the first bullet clears both; the two figures that do not are the live re-request and the cross-size parity count in the Bebop note, and both are flagged where they appear.
Every figure above is reproducible from ClearTrace's quote-sample data as published in the 2026-08-06 snapshot; the live leaderboard moves with each sync.
|
|