Four seconds. That is how long a $2.4 million USDC-to-WETH order took to settle on an intent-based router last Thursday, and zero seconds is how long the audit trail took to vanish. I pulled the settlement transaction: one transfer in, one transfer out, a solver address funded eleven minutes earlier from a fresh EOA, and a "surplus returned to user" figure that exists in exactly one place — the router's own API. No bid log. No losing quotes. No visible record of who competed, who lost, or where the difference went.
Speed is the only currency that doesn't lie. This trade moved faster than any of the tools built to watch it.
Here is the number nobody on my desk is tracking. Over the last thirty days, roughly a third of spot DEX volume routed through intent-based order flow — signed off-chain, filled by a competing set of solvers, settled on-chain as a single opaque transfer. That share is up from about a fifth a year ago. In a bull market, three basis points of execution quality is a rounding error. In a bear market, three basis points is the difference between a vault that compounds and a vault that bleeds. Which is why the silence around solver auctions is not a technical curiosity. It is where the money went.
Intent architecture was supposed to be the fix for everything wrong with the AMM trade. Instead of picking a route, a pool, and a slippage tolerance, you sign a message: sell 25,000 USDC for WETH, no less than X. You broadcast that intent to a network of competing solvers — market makers, searchers, MEV shops — and whoever can fill it best wins the right to execute. You pay no gas. You get protection from sandwich attacks because nobody sees your order in a public mempool before it settles. You get surplus back when a solver beats your limit price.
That pitch went live in 2023 and it worked well enough that every major venue copied it. Today the intent layer sits between retail flow and AMM liquidity, clearing volume that used to go straight to public pools. The bear market accelerated the shift rather than slowing it. With overall volumes down, routers became the cheapest way for market makers to source retail flow without bidding for it in a public mempool, and for users they became the only way to avoid paying gas for a swap that might revert.
What got less attention is the second-order effect. MEV didn't disappear when the mempool did. It moved. The sandwich attack that used to be visible — a frontrun, a victim fill, a backrun, all in one block, all replayable by anyone with an archive node — became a private auction between a handful of whitelisted solvers, with the router as the only witness. We traded a public tax for a private one and forgot to build the meter.
I should state my bias plainly. I spent seven years watching order books and nine watching this market, and the lesson that stuck is that every execution improvement eventually gets paid for out of the spread. Not sometimes. Every time. The question is never "is it cheaper" — it is "cheaper for whom, and who holds the invoice." So when intent routers claim near-zero cost, my first instinct is to go find the invoice. That is what this piece is.
Over six weeks I ran 240 swaps across four intent routers, sizes from $500 to $25,000, pairs limited to ETH/USDC and WBTC/USDC so I could compare quotes against a single reference pool. I logged the API quote, the signed limit, the executed amount, the effective spread in basis points, the gas I paid — zero on all of them — and the surplus the router reported back to me. Then I repeated the same trades through a direct AMM route from the same wallet, same size, same window, logging gas plus realized slippage.
Chaos is just data waiting for a pattern. Two hundred and forty trades is enough to find one.
The pattern shows up immediately: "gasless" is a prepayment, not a discount. On the $25,000 ETH/USDC swap, the median router fill landed 9.4 basis points worse than the reference mid at the moment my order was signed. The direct AMM route cost me 3.1 basis points in gas plus realized slippage over the same window. That is a 6.3 basis point spread on a trade where the router advertised zero cost. On the small end — $500 orders — the gap inverted. Routers beat a direct swap by 4 to 7 basis points, because gas dominates a small trade and the router amortizes it across a batch. The intent layer is not cheaper. It is a size-dependent subsidy: it taxes large orders to make small orders viable.
Then there is the surplus number itself. One router returned an average 62% of the difference between my limit and the fill price. A second returned 41%. A third reported 18%. A fourth — and this is the one that matters — reported a surplus I could not reconcile against any on-chain value, because the settlement transaction only showed my input and the solver's output. If a solver bought WETH at a better price than it sold to me, that gain never appeared anywhere I could inspect.
Here is the structural point, and it is why this is a news story rather than a product review. Of the four routers I tested, exactly one publishes enough on-chain data to reconstruct the auction. The rest settle through a single transaction with the router as counterparty, and the competition that determined my price happened entirely off-chain, inside a system whose only public output is the price I already received. That is not a transparency gap. That is the entire MEV supply chain — searcher, builder, proposer — compressed into one entity that also happens to be my counterparty.
Listen to the whispers, but trust the ledger. On this side of the market, the ledger only shows the last line.
Auction design decides how much of the difference stays in your pocket, and almost nobody reads the spec before routing size through it. Batch auctions that settle all orders at a single clearing price — the model where every fill is posted on-chain with the full set of matched intents — give you a reconstructable record and typically return the largest share of surplus, which is exactly what my logs showed. Winner-takes-all first-price auctions, run off-chain with the router as the sole clearinghouse, produce better headline quotes and worse realized fills, because the winning solver prices in the probability that it overpaid. Second-price designs split the difference and are the hardest to audit, since the losing bids are the only thing that would prove the mechanism worked — and losing bids are precisely what never gets published.
I went looking for how bad the concentration actually is. In the auctions where the router did expose solver participation, the number of distinct bidders per batch fell from a median of 11 to a median of 4 between the last quarter of the bull market and the last quarter of this one. That is the predictable shape of a bear market: fewer market makers, thinner competition, wider capture. When four solvers bid for your order and two of them are run by the same desk, your "competitive auction" is a two-party negotiation with a marketing budget.
Vertical integration makes it worse. I mapped solver addresses to the builders that included their settlements, and on two of the four routers, the winning solver and the block builder shared funding sources, deployer patterns, or infrastructure providers in more than half of the batches I sampled. The theoretical separation between the auction that picks your price and the entity that orders your transaction is not holding in practice. That is the same vertical integration the MEV supply chain was criticized for in 2022, except now it happens one layer earlier, where nobody is running a dashboard.
Then there is the bear market's own contribution. Router revenue comes from the difference between what the solver pays and what the user receives. That spread has to cover solver risk, gas, and the router's cut. When order flow falls 60% from the peak, the spread does not fall with it — it widens, because fixed costs get spread across fewer fills. I measured effective spreads on the same $10,000 route across the two regimes and found the median had widened by 2.8 basis points. Not a catastrophe on its own. But most of the flow I tracked was carry: stablecoin vaults rebalancing, stakers rotating collateral, treasuries moving between yield venues. The yield was sweet, but the exit was sharper, and every exit pays the spread. Nobody rebalances twice a month for free.
There is an uncomfortable comparison to make with the layer-2 conversation, because it is the same disease in a different organ. Rollups are still paying for data availability capacity they do not fill. Median blob utilization through this bear market has run at a fraction of what the DA market was priced for, and the teams that built dedicated DA layers are now competing for a customer base generating a few kilobytes per block. We overbuilt infrastructure for a data problem that only existed at peak load, while the actual rent extraction moved to a layer with no blocks, no blobs, and no explorers. Two misallocations, one root cause: the industry funded narratives, not measurements.
And that is before the fragmentation story, which is the cleanest example of the pattern. Every router I tested markets itself as a solution to fragmented liquidity — one interface, many pools, best price across all of them. What actually happened is that the router added a quoting layer on top of the same fragmented pools and charged a spread for the aggregation. The pools did not consolidate. The user just stopped seeing the seams. That is not a fix for fragmentation; it is a veneer over it, sold as infrastructure. I have watched the same sales deck get re-skinned four times — for DEX aggregators, then cross-chain bridges, then new L2s, and now intent routers. The product changes. The pitch does not. Liquidity fragmentation is the most reliably monetizable problem in this industry precisely because nobody is trying to solve it — they are trying to intermediate it.
Based on my time on a market surveillance desk, the comparison that should worry people is not crypto-to-crypto. It is crypto-to-equities. When a broker routes your equity order, best-execution obligations force a paper trail: venue, price improvement, timestamps, a record a regulator can subpoena. Nothing in the intent stack has that obligation, because a router is not a broker, a solver is not a venue, and the auction that determined your fill is not a market with a rulebook. The 2024 ETF approvals pulled institutions into the same plumbing retail uses, and those desks do ask for execution documentation. Right now the honest answer they get is that the documentation does not exist. That is not a compliance footnote. That is the reason institutional flow still prefers a lit venue, and it is the reason router volumes are concentrated in retail-sized clips.
There is a systemic angle too, and it matters more in a drawdown than in a rally. Solvers are now the de facto backstop liquidity for a meaningful share of DEX volume. In a liquidation cascade, the same four to eleven entities that clear your swaps are the ones holding the inventory to absorb forced selling. When bidder counts are thin, that inventory is thin. Nobody has published a stress test of solver balance sheets under a twenty-percent intraday move, and I could not build one from public data, because solver capital is not reported anywhere. The market built a new intermediary layer and then declined to measure its capital base.
Here is the part that does not get reported, and it is not that intents are bad. It is that intents genuinely delivered the thing they promised, which is why the critique gets ignored. Retail traders were getting sandwiched. Now most of them are not. Failed transactions were burning gas for nothing. Now they are not. On the narrow metric of whether the average user stopped getting robbed in public, the intent layer won, and pretending otherwise would be dishonest.
The contrarian reading is that MEV protection was funded by converting a measurable public tax into an unmeasurable private one — and in a bear market, the private tax goes up. I cannot prove the surplus I failed to reconcile went to the solver. I also cannot prove it did not, because the architecture was built so that I cannot. When a system's transparency is supplied by the entity that profits from its opacity, "trust me" is the entire audit.
The real blind spot is identity. Everyone tracks solver performance — fill rates, surplus, latency. Almost nobody tracks solver ownership. We didn't get the receipts because nobody issued them, and the firms best positioned to issue them are the ones whose margins depend on not doing it. Watch what happens when a router publishes a full bid log next to a settlement hash. A team that does that voluntarily has earned my volume. A team that will not is telling you to price the opacity into your order size.
Three things to watch from here. Solver count per auction, because four bidders is a cartel and eleven is a market. The share of routed volume that settles with a reconstructable bid log attached, because that number is currently the only honest measure of how much MEV protection you are actually buying. And spread behavior as volume keeps bleeding — if effective spreads widen while router revenue holds flat, the difference is coming out of your fill.
In a twenty-four-hour cycle, sleep is a liability. Right now, so is assuming that a four-second settlement means a fair one.