A fifth of Ethereum's router traffic belongs to no venue. We stopped giving it to one.

Every per-venue revert rate we publish rests on a prior step nobody reads: deciding which venue a router contract belongs to. Addresses a venue has published as its own are matched from a registry. Everything else used to be settled by fill count, winner takes all, with no margin. If two venues both routed through the same contract in a week, whichever filled more took the address, and with it every transaction and every revert recorded against it.

We measured the margin on Ethereum for the first time on September 23. The contested traffic is concentrated in a handful of addresses, and the two largest are not any venue's router. One is 0xccc88a9d, RelayApprovalProxyV3, an approval proxy operated by Relay: 37,618 fills in the scored window, with the leading venue ahead of the runner-up by 1.33x. The other is 0x4337084d, the ERC-4337 EntryPoint, the account abstraction contract every smart-account swap passes through whatever venue built the route: 13,944 fills, leader ahead by 1.04x. A third is 0x0439e60f, MetaMask's MetaBridge, 1,709 fills, lead 1.35x. At 1.04x, a quiet week on either side hands over the whole address and both of its measured populations.

So the revert query no longer picks a winner it cannot justify. Where a router's top two projects are within 2x of each other, the address is booked to a row of its own, labelled contested, and given to neither. That version deployed on four chains on September 21 and ran for the first time in the September 23 sync. On Ethereum the contested row now holds 215,897 routing transactions out of 1,131,419, 19.1% of the chain's routed flow, and it is the second-largest row in the table. It is not a venue and we do not publish it as one: it appears in the revert table and nowhere on the leaderboard.

Ethereum, 7 days	Routing txs, Sep 21	Routing txs, Sep 23	User revert rate
(contested)	did not exist	215,897	0.36%
KyberSwap	162,258	79,887	0.57% → 0.98%
Bitget	65,125	52,572	0.60% → 0.71%
1inch	134,682	128,853	0.20% → 0.18%
Uniswap	147,089	155,036	1.17% → 1.20%

Ethereum, 7-day windows ending 2026-09-21 and 2026-09-23. The all-venue total rose 2.2% between the two, 1,107,422 to 1,131,419, so the drops above are reassignment and not lost flow. User revert rate is the rate among senders whose own transactions to these routers succeeded at least 90% of the time in the window.

The honest reading of that table is that no venue in it changed behaviour between Monday and Wednesday. We did. Anyone who took a week-over-week move in these rows as a market signal would have been reading our own methodology, which is the reason to publish the change alongside the numbers rather than after them. One venue also left the Ethereum table entirely: OpenOcean sat at 1,041 routing transactions against the query's floor of 1,000, and after the carve-out it is below it, so it has no published Ethereum revert rate this week.

Two limits worth stating. The 2x margin is a choice and not a measurement, and a different threshold would draw the line somewhere else. What it is not is a threshold tuned to a result: the two largest contested addresses sit at 1.33x and 1.04x, and both moved slightly further from the line in the latest sync rather than closer to it, so neither is about to resolve on its own. And the contested row's own 1.01% revert rate is not a quality signal about anybody. It is shared infrastructure carrying flow that several venues originated, which is exactly why we no longer hand it to one of them.

Also this period

Router revert rates on Ethereum, after the carve-out. 1inch 0.18% on 123,814 user transactions, LI.FI 0.38% on 30,402, ParaSwap 0.58% on 31,022, Bitget 0.71% on 50,832, Sushi 0.97% on 38,103, KyberSwap 0.98% on 71,013, OKX 1.01% on 32,033, Uniswap 1.20% on 145,174. The highest of them is Uniswap at 1.20%. One further measured router is withheld pending its own right of reply and sits inside that range, so its absence does not move the top of it. These strip senders whose own transactions to these routers failed more than 10% of the time, which is most of the distance from a headline rate: OKX reads 5.85% before that filter against 1.01% after, KyberSwap 3.26% against 0.98%. That filter selects on the outcome being reported, so read it as the failure rate among consistently-succeeding senders rather than as the share of all attempts that fail. The comparison holds only across router venues, where the user submits the swap.

50,143 sandwich attacks in the 7 days to September 23, against $70.2M of victim volume across the chains we track. The count and the value sit on different chains: Base carried 39,525 attacks for $8.4M, Ethereum 10,067 for $61.5M, Optimism 526 for $0.2M. Across the 500 most recent victim trades in the sample the median size is $194.02, so the Base figure is a very large number of very small trades.

The quote sample now stands at 144,956 captures over 93.5 days, 123,932 of them realized in a fork simulation, 85%. 360 of 426 venue, chain and size cells are rated, meaning at least 30 realized samples spanning at least 7 days. OpenOcean's cells remain the exception on the collection side: 51% realized against 47% of captures where we obtained no quote at all, and that has been our access rather than that venue's availability since August 21.

The aggregator leaderboard, the per-router revert breakdown and the quote-accuracy sample are on the ClearTrace dashboard. The methodology notes set out how a router is assigned to a venue, what the user-only revert rate filters out, and why revert rates are comparable only across router venues.

Open Dashboard

Methodology

ClearTrace · Cross-chain DEX attribution. Revert and sandwich figures are on-chain over 7-day windows, from syncs stamped 2026-09-21 and 2026-09-23. The contested row exists only in the revert table: it does not change any attribution, volume or coverage figure, and the router assignment rule it comes from applies to routers we match by fill count, never to addresses a venue has published as its own. Contested fill counts and lead multiples are from a one-off diagnostic run on Ethereum on 2026-09-23 over the same window, so they are counted in fills while the table above is counted in routing transactions; the two are different units and are not interchangeable. The user-only revert rate removes senders whose own transactions to these routers failed more than 10% of the time, which bounds it under 10% by construction, and it is comparable only across router venues: on a batch auction or an intent venue a solver submits, so an unfilled order never becomes a reverted transaction. Quote figures come from the forward-captured quote sample, 144,956 captures between 2026-06-23 and 2026-09-24, simulated on a fork that excludes gas, mempool competition, inclusion delay and real MEV. Attribution and coverage figures are deliberately absent from this issue: that table last refreshed on 2026-09-21 and this week's story is about the revert table alone. 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.