There is a number in a preprint published this quarter that should stop every passive liquidity provider on Solana. In quiet markets, a swap routed through a professional market maker cost the trader 0.26 basis points against a reference price. The same swap, through a public automated market maker, cost 2.59. Ten times the spread, for the same asset, on the same chain, in the same second of the same block.
That gap is the entire marketing case for the professional AMM. It is also, read alongside the second number in the same study, the quiet record of a transfer. When the researchers measured markout โ what the price did two seconds after the fill โ the professional pool printed +0.37 basis points while the public pool printed โ0.22. Better execution for the trader. Positive expectancy for the professional maker. And, folded into the same table, a negative one for the depositor who never sees the trade at all.
I have spent eighteen years watching markets narrate themselves into comfort. This is the rare dataset that narrates itself into discomfort.
The numbers come from a preprint attributed to researchers at ETH Zurich and Category Labs. It is not a project announcement. Not a governance proposal. Not a token launch. It is a measurement paper, and that changes what every downstream claim is allowed to mean: the technical question here is the integrity of the method, not the promise of a protocol.
The study covers Solana, Base, and Monad. Its data window nominally runs into 2026 โ a detail I will return to, because a window that extends past the present is either a typo or a warning. More important is what the paper separates. The industry has spent three years conflating two things: execution cost, the deviation of a fill from a reference price, and markout, the favorability of the price move after the fill. The authors treat these as distinct and refuse to let either one stand in for profitability. That single act of discipline is the most valuable thing in the document.
I learned this discipline the hard way. During the 2020 DeFi Summer I tracked capital velocity between Compound and Aave and wrote an essay I titled The Yield Trap. The argument was simple and unpopular: the headline APY was a story, and the story was hiding a liquidity risk. What I did not yet have was the vocabulary to name the mechanism underneath. The vocabulary arrived later, formalized as Loss-Versus-Rebalancing.
To read this paper, you need that vocabulary. A public AMM โ Raydium, Orca, the familiar Solana pools โ quotes a price to anyone. It cannot choose its counterparty. That openness is its virtue and its cost. The cost is adverse selection: the passive liquidity provider systematically sells to arbitrageurs who know more than the pool does, and the pool rebalances at a worse price. LVR is not a bug. It is the price of permissionless liquidity, and it is conserved.
A professional AMM โ Jump's BisonFi is the cited example โ does what the public pool structurally cannot. It classifies the flow before it prices it. Inventory, quote freshness, and the quality of incoming order flow all feed the quote. Jupiter, the dominant router, runs a dedicated signer that identifies front-end traffic and tags it retail and non-toxic. The professional pool then decides, order by order, whether it wants to trade at all.
Solana's DEX landscape was built on the opposite premise. The chain's speed and low fees made the public pool the natural primitive: cheap, open, permissionless, adversarial to no one in particular. For three years that premise held. What this paper documents is the moment the premise started to invert โ not because the public pool failed, but because a better-informed participant arrived and found a way to trade around it.
The filter that produces 0.26 basis points is not subtle in principle. The researchers define quiet-flow execution as a fill where the reference price moved less than one basis point in the five seconds before and the one second after the trade. That is a proxy for non-toxic flow โ the uninformed swap, the treasury rebalance, the retail order that is not racing an oracle.
I have audited enough whitepapers to distrust proxies on sight. In 2017 I spent weeks modeling Golem's reward distribution against transaction-fee volatility and found a mechanism that looked elegant in the abstract and collapsed the moment you asked who paid for it. The lesson was not that proxies lie. It was that a proxy is a choice, and the choice is where the argument lives. Quiet flow is a reasonable definition. It is also the definition that flatters the professional pool, because the professional pool is precisely the entity allowed to select for it.
Here is what the mechanism does in practice. When flow is quiet, the professional maker quotes tight, absorbs the trade, and earns a small, positive, repeatable spread. When flow is toxic โ informed, latency-armed, aimed at a stale quote โ the professional maker does not need to quote at all, or quotes wide enough that the arbitrageur walks. The public pool has no such option. It quotes to everyone, which means it quotes to the arbitrageurs the professional pool just declined.
That is the whole trick, and it is not a trick at all. It is selection. The professional pool has privatized the right to say no.
Then comes the markout, and the markout is where the story stops being about traders.
The headline โ 0.26 versus 2.59 โ describes what the trader experiences. The markout โ +0.37 versus โ0.22 โ describes what the maker experiences. Read together, they say something the headline alone conceals: the professional pool is not earning its edge by charging quiet flow more. It is earning it by avoiding the flow that costs money and leaving that flow to the public pool.
The arithmetic is almost embarrassingly clean. A 0.59 basis point swing in markout separates the two venues. That swing is the compensation for adverse selection, and adverse selection did not disappear. It was routed. The informed order that would have hit the professional quote now hits the public quote instead, because the professional quote was wide or absent. The public pool's depositors absorb the loss the professional pool declined to take.
I want to be as careful as the authors are. Venue averages cannot prove that professional pools caused losses for aggregate passive LPs. That is their sentence, and it is correct. Proving causation requires trade-level attribution โ position, holding period, attributable revenue, attributable cost โ and over short horizons that ledger is still open. The paper says so. The paper is right to say so.
But the absence of a proof is not the presence of innocence. Math does not care about your conviction. It cares about conservation. If adverse selection is a conserved quantity, filtering it out of one venue does not delete it. It relocates it. The only remaining question is whether the party absorbing it can see the invoice.
Right now, it cannot. The passive LP sees an APR. The APR does not itemize the toxicity it swallowed.
One more methodological thread deserves pulling, because the entire headline rests on it.
Execution cost in this paper is a relative measure. Every fill is benchmarked against a reference price reconstructed from Bybit's order book microprice, converted through USDC/USDT midpoints. That is a defensible construction โ microprice weights both sides of the book by size, so it tracks true pressure better than a naive midpoint. It is also a single external reference. If Bybit's book is thin, skewed, or momentarily dislocated from Solana's own liquidity, the benchmark moves, and every execution-cost number moves with it.
This is not a flaw unique to this study. It is a structural condition of measuring DeFi against CeFi. The reference venue is a centralized exchange; the measured venue is a chain; the bridge between them is an assumption. I learned to name that assumption out loud during the 2024 ETF cycle, when I wrote a report I called The Boring Boom. The thesis was that volatility would fall as narratives standardized around regulatory clarity โ and that understanding the story behind the regulation would be worth more than the regulation itself. The same logic applies here. Understanding the story behind the reference price matters more than the reference price.
So when you read 0.26 versus 2.59, read it as a comparison within one reference frame, not as an absolute cost. The frame is probably right. The frame is not self-evident.
Move up the stack, because the venue is not the whole system. The router is.
Jupiter does not merely aggregate. It classifies. Its dedicated signer marks front-end flow as retail and non-toxic, and that label travels with the order into every venue that can read it. A professional pool that can see the label has been handed a sorted deck. The public pool, which prices against state rather than provenance, receives whatever the sorting leaves behind.
This is not a conspiracy. It is an incentive gradient, and the gradient runs one way. Every venue wants the good flow. The venue that can identify it first will win it. The router that can label it first becomes the most valuable chokepoint in the chain โ more valuable than any single pool, because it decides which pool sees what.
That is a new kind of centralization, and it does not announce itself. It hides behind the word routing. Narratives are liquid; truth is solid, and the solid truth here is that a label is a pricing decision. The moment retail and non-toxic became a tag, it became an asset.
I have watched this film before, in traditional markets. Payment for order flow, best-execution obligations, fair-access rules โ the entire architecture of US equity microstructure exists because someone eventually asked who was allowed to see the order before it printed. DeFi has rebuilt that architecture minus the disclosure. The tag retail and non-toxic is a client-tiering decision wearing a technical costume.
Then there is the number that should worry every router. On Base, a venue called Tessera showed an execution price that differed from its own reconstructed prior-block-end quote by 1.08 basis points on a trade-count average, and 0.56 by volume. The researchers call the block-timing pattern spoofing, and then do something rare โ they state plainly that there is no direct evidence of operator intent.
Hold both facts at once. Better execution and worse quoting can coexist, because they are measured against different things. A trader can receive a great price and still be trading against a quote that was never real. The reference moved. The fill did not follow. The difference is a toll paid by whoever trusted the advertisement.
The router 0x responded the way a competent risk desk responds: it said it would cut the source until the execution problem was resolved. That single sentence is more consequential than any governance vote I have read this year. A router voluntarily enforcing an execution standard is the first rough draft of a DeFi best-execution norm โ written not by a regulator, but by the party whose reputation is on the line every time it routes a fill.
And there is a gap even the authors leave open. Their parity test checks whether quote and execution share the same state snapshot, but only at a single snapshot. Cross-block persistence โ whether the discrepancy recurs or was a one-off โ remains unmeasured. A single snapshot is a photograph. Persistence is a film. The industry is currently arguing about a photograph.
If the sorted deck keeps sorting, the end state is predictable. The public pool becomes the venue of last resort for flow that no one else wants. Its LP base absorbs the residual toxicity the professional pools declined. Yield compresses. Compressed yield triggers withdrawal. Withdrawal thins the book, which widens the spread, which makes the public pool even less attractive for quiet flow โ which sends more of it to the professional venue. The loop is self-reinforcing, and it does not require anyone to behave badly. It only requires everyone to behave rationally.
I call this a degradation equilibrium, and I have seen its emotional signature before. In 2022, after Terra and Luna, I retreated to a cabin outside Austin for three weeks. I could not read the discourse; the volume of broken trust was too high. What I did in that silence was reconstruct the failures of Celsius and BlockFi and discover that the word decentralization had been a facade over a balance sheet. Solitude is the price of clear vision, and the clarity it bought was this: a system can look distributed and still concentrate risk in one unexamined place.
The propAMM is that unexamined place now. It is not a fraud. It is a filter. And a filter, by construction, has a clean side and a dirty side. The clean side is priced beautifully. The dirty side is where the passive LP has been quietly seated, holding a ticket to a table it did not choose.
Here is the part that keeps me up. The loss is not merely unmeasured. It is unmeasurable by the people absorbing it. A passive LP cannot see which of its fills were toxic; that information lives with the router and the professional maker. The party bearing the cost is the party with the least visibility into it. This is the structural asymmetry the paper exposes, and it is more durable than any single dataset. You cannot manage a risk you cannot observe, and you cannot observe a risk that someone else has been given a signer to label.
There is a version of this future that is worse than compression. The paper hints at it when it describes private market-making logic hidden behind public settlement. If a venue can show one price to the router and execute another against a specific counterparty, you no longer have an AMM. You have a shadow AMM โ a public facade over a private book. The label would be the same. The fills would not.
Strip the mechanism to its commercial core and the competitive map is unambiguous.
The professional pool wins. It sees the flow, filters it, and keeps the clean side. The router wins, because complexity is its moat โ the harder it is to route well, the more valuable a router becomes. The passive depositor loses, not by theft but by seating. And the public AMM โ Raydium, Orca, the whole generation of permissionless venues โ faces a choice it has never had to face: either adopt its own flow-quality logic, which blurs the line between public and private liquidity, or accept the role of residual absorber for the flow nobody else wants.
That is the real competition of this cycle. Not fees. Not TVL. Flow provenance. The venue that can tell a clean order from a dirty one before it prints has an edge that compounds, and the venue that cannot is condemned to price against everyone.
Jupiter sits in the most uncomfortable chair of all. Its signer hands professional pools a sorted deck, which makes it indispensable to them and suspect to everyone else. Its routing logic decides which pool sees what, which makes it the arbiter of the entire structure โ and arbitration without a published standard is not neutrality. It is discretion. A router that grades order flow is a market operator, whether or not it files as one.
And the sorting is about to get worse. In my current work on the AI-and-blockchain convergence, I keep arriving at the same uncomfortable conclusion: autonomous agents will be the most latency-sensitive, most information-advantaged flow that has ever hit a chain. An AI agent that can read the reference book, model the pool's staleness, and route its own order is the definition of toxic flow โ perfectly informed, perfectly fast, perfectly indifferent. When that flow arrives at scale, the filter will not be a competitive advantage. It will be a survival requirement. And the public pool, which cannot filter, will be the venue of last resort for the machines that no one else will quote.
The consensus reading โ professional makers are extracting from passive LPs โ is emotionally satisfying and analytically premature.
Consider the alternative. If propAMMs did not exist, the toxic flow would not vanish. It would hit the public pools anyway, because there would be nowhere else for it to go. The passive LP's loss under a single public venue is the full LVR. Under a two-venue system, the passive LP's loss is the LVR the professional pool declined. Those two quantities may be close to equal. The professional pool may be capturing a margin that was always latent, not manufacturing a new one.
What is genuinely new is not the extraction. It is the visibility. Before the labeling, the transfer was anonymous โ it appeared as a slightly worse APR and no one could name the counterparty. Now the counterparty has a name, a signer, and a tag. The loss may not have grown. It became legible.
That distinction changes what you should do. If the problem is extraction, you flee public pools. If the problem is legibility, you do something harder: you demand that every venue disclose its flow-quality assumptions and you price liquidity accordingly. In the chaos, look for the invariant โ and the invariant is that adverse selection is conserved while disclosure is optional. The crowd sees a tenfold improvement in execution. I see a sorted deck and an unoccupied losing seat.
Then there is the word the authors chose and the word they disowned.
They call the Base pattern spoofing. Then they state there is no evidence of intent. In a legal context those two sentences cannot both be load-bearing. Spoofing is a defined offense with a defined intent requirement, and the CFTC has pursued it aggressively in traditional markets. A researcher using the word descriptively and a regulator using it enforcement-first are not speaking the same language, and the gap between them is where liability lives.
I do not think a regulatory action is imminent. I think something subtler is happening. The DeFi equivalents of best-execution obligations, order-flow-payment disputes, and fair-access rules are quietly being drafted โ not in statute, but in router policy. 0x cutting a source until execution is fixed is soft law. Jupiter's tagging is soft law. A preprint from ETH Zurich is soft law. Hard regulation usually arrives to ratify a standard the industry has already adopted, and the industry is currently adopting one without noticing.
Watch for the moment a router publishes its flow-classification methodology. That document, not any token, will be the most important thing written in DeFi this cycle. It will be the first page of the rulebook.
So watch the routers, not the pools. The next structural fight on Solana will not be fought over fees or TVL. It will be fought over whether a venue may see the provenance of an order before it decides to trade. Whoever writes the disclosure standard โ a router, a preprint, or eventually a regulator โ will set the price of liquidity for the next cycle. The 2024 ETF approval taught me that the boring institutional narrative always wins the long game; this is the DeFi version of that lesson, arriving early. The ledger will settle. The only question is who is still seated when it does.