Clara Bennett: You look like someone who just read something annoying — what happened?
Finn Brooks: Okay, not annoying, more like — confusing in a way I can't shake. I've been staring at this data point about Solana and I need you to tell me if I'm reading it wrong.
Clara Bennett: Let's hear it.
Finn Brooks: Proprietary AMMs — decentralized exchange protocols on Solana — have been quoting better prices than Binance. Better bid-ask spreads. Tighter than a top-tier centralized exchange.
Clara Bennett: Mm — and the bid-ask spread is the core efficiency metric. The tighter it is, the less the trader loses on execution.
Finn Brooks: Exactly, so if DEXes are winning on that — the thing they were never supposed to win on — shouldn't that be the headline that ends the whole conversation?
Clara Bennett: It would be, except — how is the price being set?
Finn Brooks: That's the thing. These proprietary AMMs — propAMMs — their quoting logic is closed-source. Like, genuinely opaque. No one outside the operator knows the actual mechanism.
Clara Bennett: So the decentralized exchange is beating Binance on price, but we can't see how.
Finn Brooks: Which is — I mean, that's either a triumph or a trick, and I genuinely don't know which one yet. Because the whole point of a DEX is censorship resistance — no operator in control — and a black-box pricer kind of... quietly hands that control back to someone.
Clara Bennett: That black-box point is actually where I want to start — because before we can even judge whether the propAMM thing is a triumph or a trick, we need a cleaner picture of what a DEX is actually doing under the hood. And I think the simplest way in is an analogy.
Finn Brooks: Yeah, please. Give me the non-jargon version.
Clara Bennett: A centralized exchange — Binance, let's say — is one fast auctioneer standing in one room who can see every buyer's price and every seller's price at the same instant. A DEX is that same auction running across thousands of computers that all have to agree on every single bid before anything actually executes.
Finn Brooks: Okay but wait — can't you just make the computers faster? Like, isn't that what Solana is?
Clara Bennett: That's the instinct everyone has, and it's — it's not wrong, but it misses the actual bottleneck. Speed is one constraint. The deeper issue is that distributed consensus and on-chain settlement aren't just slow — they structurally fragment where the liquidity sits. Binance runs one matching engine. One. That's why its order books are deep and its spreads are tight. A DEX, by definition, doesn't have a trusted operator running that engine.
Finn Brooks: So the spread isn't wide because DEXes are laggy. It's wide because there's no one coordinating.
Clara Bennett: Exactly — it's architecture, not lag. Now, the dominant DEX model solved part of this around 2018 with the Automated Market Maker. Instead of matching discrete buy and sell orders, an AMM prices assets algorithmically out of a liquidity pool. Uniswap made this famous — the formula is literally x times y equals a constant. No order book, no matching engine.
Finn Brooks: Which is brilliant for bootstrapping, but — hang on — doesn't that just mean the pool is always slightly mispriced versus the real market?
Clara Bennett: That's the impermanent loss problem, and we'll land there. But stay with the structure for one more second. Some DEXes tried to go the other direction — build an on-chain limit order book, actually store and match discrete orders on the blockchain the way a CEX does. MultiversX ran experiments on this. The result? Computationally brutal. Slow and expensive.
Finn Brooks: So the most CEX-like DEX design is also the most broken one.
Clara Bennett: In practice, yes. And that's the structural trap — the same properties that give you censorship resistance, no trusted operator, on-chain settlement, distributed consensus, are the exact same properties that widen spreads and slow execution. It's not a bug you can patch. The tradeoff is baked into the architecture.
Finn Brooks: When a propAMM on Solana quotes tighter than Binance, it's not solving that tradeoff — it's sidestepping it by putting a private operator back in the loop.
Clara Bennett: Right — and that's the question worth sitting with. If the only way to beat Binance on spreads is to quietly reintroduce a trusted intermediary with opaque pricing logic, then what exactly got decentralized?
Finn Brooks: But wait — actually, before we get to what the propAMMs are doing to that tradeoff, I keep getting stuck on the moment the tradeoff got created in the first place. Like, 2018, AMMs show up — and I want to make sure I actually understand why that was such a big deal structurally.
Clara Bennett: The chicken-and-egg crisis. Before AMMs, a DEX needed deep order books to attract traders, but needed traders to build deep order books. Nobody shows up to an empty market.
Finn Brooks: And AMMs just... cut that knot entirely. Because Uniswap doesn't need orders. It needs a pool.
Clara Bennett: Right — anyone can deposit assets, and the pool executes trades algorithmically. The price isn't matched from bids. It falls out of the ratio of reserves. x times y equals k. You change x, k stays fixed, y adjusts, that's your price.
Finn Brooks: Okay but that's the part that unsettles me, because — hang on — if the price is just a ratio, then the pool is almost always slightly wrong versus whatever ETH is actually trading at on Binance. Which means somebody is going to come in and correct it.
Clara Bennett: Arbitrageurs. And that's exactly who Capponi and Jia traced in their 2021 work. They formally proved that when external market prices move, arbitrageurs rebalance the pool — and they do it at the liquidity provider's expense.
Finn Brooks: So the person who deposited to make the pool work is the one eating the loss. That's — I mean, that's a brutal design.
Finn Brooks: Picture someone — let's say it's a Saturday night, she deposits equal ETH and USDC into a Uniswap pool, earns fees all week, logs back in the following Saturday. ETH moved sharply. And instead of holding what she put in, the pool got drained toward whichever token got cheap — by arbitrageurs who pocketed the spread. She holds less total value than if she'd just... done nothing.
Clara Bennett: And Capponi and Jia's model specifies when it's worst — high exchange rate volatility, low correlation between the token pair. Those two conditions together maximize what arbitrageurs can extract.
Finn Brooks: So impermanent loss isn't just a risk you're warned about in the fine print. It's load-bearing. It's in the x times y equals k formula itself.
Clara Bennett: Now, you can tune the curve — higher curvature in a CFMM does reduce how much arbitrageurs can extract. But the tradeoff flips: protect providers more and you hurt trader efficiency. There's no free parameter that makes both sides happy simultaneously.
Finn Brooks: Which is — okay, that's the part that I find genuinely unsettling, because it means every design choice is just deciding who absorbs the cost. Not eliminating it.
Finn Brooks: And honestly, that makes what comes next even messier — because when propAMMs on Solana and Layer 2 hybrids try to narrow that gap, they do close it, but only by pulling in sequencers or opaque operators, and I think that story is worse than it looks from the outside.
Clara Bennett: Worse than it looks is actually the right frame — because the dYdX case is where that pattern gets most uncomfortable.
Finn Brooks: Wait — dYdX isn't even an AMM. That's an order book.
Clara Bennett: Exactly. Which is why it looked like the clean solution — familiar mechanics, no constant-function pricing, no impermanent loss issue. But the matching still has to happen somewhere, and in dYdX's design, that somewhere is off-chain. Off-chain relayers. Which means you've got a DEX that settles on-chain but coordinates off it.
Finn Brooks: So the operator function just... moved. It didn't disappear.
Clara Bennett: That's the consistent pattern. And Layer 2 rollups do the same thing more subtly. Ethereum Layer 2s batch transactions off the main chain, settle proofs on-chain — latency drops, gas drops, and suddenly medium-frequency trading strategies that were impossible on Ethereum mainnet become viable. That's real progress.
Finn Brooks: But the sequencer.
Clara Bennett: The sequencer. That's the piece — I mean, the sequencer is technically one node that decides transaction order before anything gets batched and settled. It's split across a few components on paper, but in practice those components still aggregate to a single point of control. That is a trusted intermediary. Under a different name.
Finn Brooks: No but that's genuinely kind of ironic, right? Like you built the whole architecture to not have a Binance, and then you have a sequencer that has basically the same power a Binance operator has over your transaction.
Clara Bennett: Now picture a trader — say someone's running a small fund from a home office in Lisbon, she's built a strategy around a Solana propAMM because the spreads are genuinely tighter than what she'd get on Binance. She doesn't know the pricing logic because it's closed-source. She's assuming the operator isn't front-running her orders because, well — who would she even complain to?
Finn Brooks: And that's the Solana propAMM case exactly — better spreads, private operator updating quotes continuously, no fixed on-chain invariant. It's not decentralized pricing. It's just a smaller, less regulated version of a centralized market maker.
Clara Bennett: And Solana's throughput is what makes it possible at all — those continuous quote updates require high-frequency mechanics that would choke on a slower chain. So the result isn't general. It's chain-specific. You can't lift it and drop it on Ethereum mainnet.
Finn Brooks: Which means the 'DEXes are catching up' story actually has a footnote: on one specific chain, with a specific design, by handing pricing control to someone you can't audit.
Clara Bennett: And Aoyagi and Ito's model in the Journal of Political Economy Microeconomics makes it stranger still — they show arbitrageurs are already linking CEX limit order books and DEX CFMMs in real time. The venues aren't competing in isolation. Which means even if you never touch a DEX, DEX pricing dynamics are already inside the CEX price you're seeing.
Finn Brooks: Wait, so the entanglement is already there whether you want it or not.
Clara Bennett: Right — and that's what makes the pattern so worth naming. Every mechanism that closes the spread gap introduces a new dependency or trust assumption somewhere. propAMMs: opaque operator. Layer 2s: sequencer. dYdX order books: off-chain relayers. The efficiency problem gets quieter. The operator function just changes its address.
Finn Brooks: And then you hit the question underneath all of it — like, who is actually paying for this? Every time someone uses a DEX and gets a slightly worse fill than they'd have gotten on Binance, that cost is real. But most of those users... they're probably never going to face censorship. They're not dissidents, they're not sanctioned entities, they just — they want cheaper gas fees or permissionless access or whatever. So they're absorbing the efficiency tax for an architecture that was designed to protect people in a completely different situation.
Clara Bennett: That cross-subsidy almost never gets made explicit. A retail trader on Uniswap isn't told: your wider spread is partially funding a system that lets someone in a restricted jurisdiction trade without getting blocked. That's the actual deal. And the Capponi and Jia impermanent loss work makes it even more layered — even the liquidity providers bearing that loss mostly aren't the ones who need censorship resistance most urgently.
Finn Brooks: Right — but the part that doesn't fit is that DEX volumes kept growing even when the efficiency gap was obvious. Like, Aoyagi and Ito's whole framework assumes rational adoption under specific conditions, and yet the empirical spread is much broader than that. So either people are... irrational, or — wait, actually — they're buying something the model isn't pricing. Permissionless access. Just the option to not need permission.
Clara Bennett: The option value of being able to transact without asking. Yes. Even if you never use it.
Finn Brooks: Which is a weird thing to land on because — I came in this morning genuinely confused about a data point. A DEX quoting tighter than Binance. And the answer turned out to be: that's not the DEX winning. That's the DEX becoming what it was supposed to replace, just smaller and less regulated. The tension isn't a bug to patch. It's the design choice. Every efficiency gain buys a new form of opacity or trust somewhere, and that's — I mean, that's probably just permanent.
Clara Bennett: The next generation of DEXes won't be asking whether they can match CEX efficiency. They'll be asking whether they can do it while staying genuinely decentralized. And every time they've gotten close so far — propAMMs, Layer 2 sequencers, dYdX's off-chain relayers — they've borrowed a little bit of the center to get there.
Finn Brooks: You came in this morning looking like someone who read something annoying.
Clara Bennett: And now you look like someone who figured out why it was annoying. That's probably a better place to end up.
