Engineering the Global Financial Infrastructure with Max Resnick | EP 162
Max Resnick’s Constellation proposal aims to make Solana a global “decentralized Nasdaq” without turning its 800-validator network into a worse centralized exchange. Co-locating every node would sacrifice the premise; multiple concurrent proposers instead create localized ingestion and competition. Resnick calls this constrained optimization: “I’m not God,” and the protocol must survive validators acting in their own economic interest.
Constellation’s headline mechanism is a 50-ms protocol-enforced economic tick across a network where New York–Tokyo round-trip latency is roughly 200 ms. Attesters use synchronized local clocks to certify which transaction shreds arrived before each batch deadline, preventing late submissions from entering earlier batches. Resnick describes operating below the network round trip as “breaking the sound barrier.”
The research began with the realization that removing a single leader’s power to reorder transactions merely shifts that power into censorship. That makes single-leader chains structurally poor at auctions, a category covering much of finance. Years of iterations—from Multiplicity through Braid, Supernova, Spectral, and now Constellation—reduced latency and bandwidth overhead while preserving safety and liveness.
The economic prize is vastly larger than crypto’s current asset base: Resnick sizes global securities at about $500 trillion versus $3–4 trillion for crypto. His conditional vision is for securities to be issued, traded, and eventually settled on Solana without leaving the chain, replacing layers of brokers, internalizers, custodians, transfer agents, and DTCC bookkeeping. “It’s certainly not a 100% chance,” but he believes regulatory moves now exist that could get the market closer.
A badly routed $50 million trade involving Aave’s aUSDT deposit token illustrates how single-leader infrastructure accretes fee-taking middlemen. The front end, CoW Swap auction, winning searcher, another backrun auction, Titan’s internal auction, relay, builder, and validator all touched the flow; the searcher paid about $30 million to Titan while keeping $10 million. Resnick’s point is not that participants are evil: “Somebody leaves some money on the table, then somebody’s going to go pick it up.”
MCP is deliberately a toolbox, not a mandated exchange design. Applications could test dual-flow batch auctions, proprietary-market-maker competition, or continuous flow trading and let performance decide the winner. Resnick thinks prop AMMs may fit on-chain markets better than order books, but their “dark side” requires routers such as Jupiter, Titan, DFlow, and OKX to become smarter.
Revenue is a protocol constraint, not an afterthought: first-come-first-served ordering could erase most validator revenue because Solana currently earns primarily from priority fees. Base fees are only about 0.04 cents per transaction, so Resnick rejects “nuking” the revenue stream of a roughly $50 billion asset for an idealized design. The objective is discounted future cash flow—“grow the pie” now, then tune extraction somewhere between 0% and 100% without driving applications away.
1. Single-leader ordering turned out to be a censorship problem
Resnick dates his path to roughly 2022–2023, during the FTX crash, when he began studying MEV from “square one” without knowing PBS, Flashbots, or Solana’s block-builder machinery. His first-principles question was simple: why should a leader choose transaction order at all?
Putting the ordering rule into the protocol removes discretionary reordering but leaves the leader able to censor. That became the load-bearing insight behind “Censorship Resistance in On-Chain Auctions”: a single-leader blockchain cannot conduct auctions particularly well, and “a lot of things that we do in finance look like an auction.”
The answer was potentially more than one leader. The work progressed through Multiplicity—the parent protocol of Fossil, which is discussed on Ethereum—Braid, an unnamed multiple-concurrent-proposer design, Supernova, Spectral, and finally Constellation, repeatedly cutting bandwidth and consensus rounds until the team believed implementation was practical.
Logan Jastremski’s market-structure framing is deliberately contrarian: “99 market makers out of 100” would reject the design because all they know is co-location. His distinction is co-location versus localized ingestion—a venue capable of receiving information near where it originates, rather than forcing the world through one regional data center.
2. A decentralized Nasdaq cannot be built by co-locating the world
Solana already has about 800 validators and an existing application ecosystem. Resnick therefore rejects “put all the nodes in one place and make a slightly worse version of the centralized exchange”; that may compete as a centralized venue, but it is incompatible with the network Solana claims to be.
His design method is “constrained optimization.” Traders may want globally instantaneous information and perfectly fair first-come-first-served ordering, but protocol designers cannot dictate every validator’s software: “If they can make money running something else, then they’re going to do it.”
Jastremski preserves Anatoly Yakovenko’s boat analogy: if a boat carrying iPhones travels down to Singapore, a nearby trader should act before that information traverses fiber to New Jersey. Multiple proposers turn geography into competing local ingestion points rather than treating distance solely as a handicap.
Resnick also rejects treating arbitrage as a universal constant. It arises when one party quotes a price another wants to take, can be modeled, and creates deadweight loss that market structure might reduce. His analogy is the moon landing: progress came from mathematics and causal models, not intuition inherited from “building wheels for wagons.”
3. Today’s financial plumbing monetizes layers of path dependence
Resnick’s sharpest example is a roughly $50 million trade from Aave’s aUSDT deposit token into AAVE. Bad routing created an enormous backrun that passed through a front-end fee, CoW Swap, a winning searcher, another backrun auction, Titan’s internal auction, a relay, and the validator; the chain resembled “10 intermediaries all taking a fee.”
Jastremski says that experience made him “super bearish block builders.” Resnick’s qualification matters: builders and landing services naturally emerge from single-leader deficiencies. The behavior is harmful to users, but it does not require evil actors—unclaimed value will be captured when incentives permit it.
He reads Ethereum’s PBS history as defensive engineering for an “infinite garden” of non-trading applications such as Farcaster. Its designers viewed validator-searcher integration primarily as a centralization threat to those applications, not as a trading flaw; Solana, by contrast, began with an on-chain order-book and “decentralized Nasdaq” ambition.
Traditional securities are similarly layered: Robinhood may route to an internalizer such as Citadel or Virtu, after which transfer agents, street-name custodians such as BNY Mellon, beneficial-ownership records, and DTCC complete settlement. Resnick says changing an entry in DTCC’s ledger alone costs six cents; blockchain settlement could move the trajectory from T+5, T+2, or T+1 toward “T+0.”
4. The addressable market extends far beyond native crypto assets
Resnick estimates global securities at $500 trillion, about half in the US and roughly half of that again in US equities, versus only $3–4 trillion across crypto assets. His goal is issuance, trading, and settlement on Solana “without ever leaving the chain.”
The regulatory opening is real but explicitly conditional: “It certainly is not a 100% chance that we can get there.” He nevertheless sees “moves on the chessboard” in the US and abroad that could permit on-chain securities if the ecosystem makes the right technical and policy choices.
Resnick describes NFTs and meme coins as stress and liveness tests for broader markets—equities, FX, derivatives, and eventually trillions in daily volume. Trading applications already demonstrate the distribution opportunity: Jastremski cites Axiom reaching hundreds of millions in revenue through a differentiated front end connected to shared liquidity.
Resnick remains surprised by meme-coin demand, but treats it as evidence that people want to trade. A few hundred thousand users may trade “a fake Pepe coin”; the capacity changes if nearly every American who owns stocks can instead trade Apple, Nvidia, SpaceX, or other securities they trust with meaningful wealth.
5. Constellation makes 50 milliseconds an enforceable economic boundary
Constellation targets “selective censorship resistance” without sacrificing consensus safety or liveness. The design discussion targets reducing prospective bandwidth overhead from roughly 20x to 5x and now about 2x, while closing operational issues such as fee-payer denial-of-service that matter when code replaces whiteboard symbols.
Its headline is a 50-ms protocol-enforced economic tick. With an approximately 200-ms New York–Tokyo round trip, Resnick argues this would be the fastest protocol of its kind on a genuinely global network: “There’s a kind of a sound barrier that we’ve been able to pass.”
The protocol does not timestamp every trade. It gives batches deadlines so a proposer cannot withhold a transaction and insert it into an earlier batch; attesters run approximately synchronized local clocks and report every 50 ms which block shreds arrived in time. Small clock errors shorten or extend the window slightly rather than breaking consensus.
Multiple proposers produce transaction shreds that are aggregated and ordered using priority fees. Interleaved attestations are designed so one attester—or even f attesters—cannot manipulate inclusion, while the surrounding machinery prevents leaders from submitting late or selectively censoring flow.
6. Fifty milliseconds is a starting point, not a sacred constant
Resnick contrasts the choice with the 2014 Budish batch-auction paper, whose 200-ms interval was not justified and whose model assumed information reached everyone simultaneously. A 2021 follow-on introduced fast and slow participants, jitter, and staggered information, showing that optimal batch length depends on the asset—SOL and the S&P should not necessarily share one setting.
Today, Resnick estimates a 10–30-ms landing advantage between sophisticated and unsophisticated senders. A 20-ms batch would therefore be premature, while ±1-ms clock drift consumes a much smaller share of a 50-ms window; better synchronization and sender tooling should eventually permit shorter ticks.
The change is less alien to current infrastructure than it sounds: Solana already disseminates shreds at a modal interval of about 32 ms. As users become unwilling to tolerate even 25 ms of avoidable delay when targeting a 50-ms boundary, competitive pressure should narrow landing disparities.
The paper models 16 proposers, though deployment could begin with four and grow to 16. Few parameters depend directly on proposer count beyond the bandwidth needed for additional transactions; the longer-run endpoint could let every validator propose, with bandwidth proportional to stake.
7. MCP lets applications compete while the protocol protects revenue
Resnick’s framing is intentionally application-neutral: “The idea for me of MCP is not what exact design is the best for the application,” but how to supply a toolbox. Dual-flow batch auctions, prop-MM competition, and continuous flow trading can all be built on it until actual usage reveals which design works.
Prop AMMs were not predictable in 2022, but their organic emergence signals a living ecosystem. Their growing pains come partly from routers designed for “dumb AMMs”; Jupiter, Titan, DFlow, OKX, and peers must adapt routing logic for market makers that discriminate among flows, pools, and potentially toxic traders.
Resnick’s strongest economic belief is that “the chain should make money.” The right relationship resembles AWS: applications should begrudgingly accept infrastructure fees, but extraction cannot become so severe that they leave. Ethereum’s post-DeFi-summer experience is his warning against maximizing one year’s revenue at the expense of future activity.
Prop AMMs expose a fee imbalance: an oracle update or cancel may use roughly 42 compute units versus about 200,000 for a trade, giving updates an enormous scheduler advantage. Resnick wants them perhaps 50x cheaper than taker flow, not 5,000x cheaper; possible knobs include a 2,000–5,000-CU scheduler minimum or base fees tied to executed compute.
Jastremski notes exchanges routinely charge by volume—roughly two basis points at Binance VIP tiers and perhaps four or five at Hyperliquid, with his figures explicitly offered for correction. Solana cannot easily copy that model: a $100,000 SOL transfer and a $1 transfer pay the same fees under the current design, while value-based charges can be bypassed by transferring ownership of an account.
The internal dispute over first-come-first-served ordering therefore collided with economics. Solana’s approximately 0.04-cent base fee is negligible, while priority fees fund validators; Resnick will not “nuke the revenue stream of a $50 billion asset to zero overnight” for an ordering rule nobody yet knows how to implement robustly.
His desired dial is neither 0% nor 100% extraction but something like 10%, discovered over time. Solana remains a growth business, so current revenue can be sacrificed for a larger future market: “It’s expected future revenue on a long horizon. Discounted future cash flows.”