The Empty Field
In the spring of 2025, I sat in a glass-walled room in Downtown Brooklyn with four people who manage a nine-figure digital-asset book, and I watched a spreadsheet fail to answer a simple question.
The spreadsheet had forty-seven rows. It was the kind of document institutional allocators build when they want to feel rigorous: token unlock schedules, sequencer revenue, monthly active developers, time-to-finality, bridge TVL, a column for "team," a column for "backers," a column for "narrative." Every row was populated. Every row had a number attached to it, and every number had a source.
Except two. Row thirty-one asked who holds the upgrade keys. Row thirty-three asked whether the entity that governs the protocol has any legal personality at all. Both were blank.
Nobody in the room wanted to talk about them. The conversation kept sliding back to throughput โ to the fact that the chain in question had just shaved another few cents off a swap, that its fee curve had crossed below the chain next door, that a payments company had quietly started routing settlement traffic through it. All of that was true. All of it mattered. And none of it answered the two blank rows, which were, in the end, the only rows that would determine whether the thing survives a subpoena, a hostile fork, or a coordinated exit.
I have been in this industry long enough to recognize the pattern. The questions a market refuses to ask during its expansion phase are precisely the questions that define its contraction. The blank cell is not an oversight. It is an avoidance, and avoidance at scale is a structural property, not a personal failing.
So I want to spend the length of this essay on two blank fields, because they sit at the center of what I have come to believe is the real Layer 2 conflict of this cycle โ a conflict that has almost nothing to do with the proving systems everyone argues about, and almost everything to do with who can persuade the most chains to join a club.
Conscience over consensus. The consensus right now is that this is a technology race. My conscience, and my audit logs, say otherwise.
How a Roadmap Became a Land Rush
To understand why the argument matters, you have to trace how we got here, because the current shape of the Layer 2 market was not inevitable. It was chosen โ several times, by several different groups, for reasons that were only partly technical.
Ethereum's rollup-centric roadmap was a bet made under constraint. Around 2019 and 2020, the base layer was congested, the fee market was brutal, and the research community had largely concluded that execution would have to move off-chain while settlement and data availability stayed anchored to Ethereum. The phrase that circulated was "rollups as the only scalable path." It was a reasonable conclusion given the tools available. It also created a structural vacuum: if execution moves to many chains, someone has to define how those chains relate to each other, and that definition is inherently political.
For the first few years, the market treated Layer 2s as a single category. Arbitrum, Optimism, zkSync, StarkNet, and later Base competed on the same axes: cheaper gas, faster confirmations, more generous incentives, better developer documentation. Token launches and points programs pulled users across bridges in waves. TVL became the scoreboard, and bridges became the scoreboard's plumbing.
The Dencun upgrade in March 2024 broke that scoreboard. By introducing blob-carrying transactions through EIP-4844, Ethereum gave rollups a dedicated, cheap data channel, and the cost of posting data collapsed. Fees on the major rollups fell by an order of magnitude within weeks. What had been the dominant cost line for a rollup โ calldata โ became a rounding error, and then, after the Pectra upgrade raised blob capacity further in May 2025, it became a rounding error twice over. Fusaka, with PeerDAS and another blob increase, pushed the same curve further out.
That should have been a moment of clarity. Instead it was a moment of panic, because it destroyed the cheapest differentiator Layer 2s had.
When data availability stops being the bottleneck, what remains? Proof generation, which is converging. Sequencer hardware, which is commoditizing. Prover costs, which fall with every generation of GPU and every algorithmic improvement. In other words, the things that made an individual rollup feel expensive to run stopped being the things that separated it from its neighbors.
Once the cost curve flattens, differentiation migrates from engineering to business development. That is not a cynical observation; it is an accounting one. If your product becomes cheaper to replicate, the only durable asset you have left is the set of counterparties who have already integrated with you.
And so, somewhere between the blob fee collapse and the institutional inflows that followed the ETF era, the Layer 2 conversation quietly changed subject. It stopped being about proofs and started being about stacks: OP Stack, ZK Stack, Orbit, the Polygon CDK and its AggLayer, Starknet's stack offering. It stopped being about individual chains and started being about federations of chains.
This is the context in which I want to examine two claims. The first is the claim everyone repeats: that this is a competition between optimistic and zero-knowledge proving systems, and that the technically superior approach will win. The second is the claim almost nobody makes out loud: that the competition is really a distribution contest, and that the winning stack will be the one whose governance and legal architecture is easiest for other people to adopt.
I think the first claim is mostly dead. I think the second claim is where the entire cycle is being decided, and I think the industry is dangerously unprepared for the consequences of the model that is winning.
What the Two Stacks Are Actually Selling
Strip the marketing away and the two dominant stack offerings describe remarkably similar products.
OP Stack gives you a rollup you can deploy in a weekend, hooked into a shared bridge specification, a shared upgrade cadence, and a governance process that lives partly on-chain and partly in a foundation. ZK Stack gives you a rollup with a zero-knowledge prover, a shared interop layer, and a governance model built around a token that exists to pay for proving and to coordinate upgrades. Arbitrum's Orbit gives you an L3 that settles to an L2. Polygon's stack gives you a CDK chain plugged into an aggregation layer.
The technical differences are real and they matter at the margins. Zero-knowledge proofs give you faster finality to Ethereum and a cryptographic rather than economic security assumption. Optimistic systems give you cheaper proving and, historically, a simpler path to production. STARKs versus SNARKs, recursive proofs, GPU provers versus FPGAs โ the engineering literature on this is serious and I have spent many nights inside it.
But here is what I learned after four months of auditing a fundraising platform's contracts in 2017 and then a decade of watching protocol choices get made: the technical argument is almost never the deciding vote. It is the argument people use to justify a decision they have already made for other reasons.
So let me be specific about what the stacks are actually selling, in the order that buyers care.
They are selling low switching costs. A team that wants its own chain has three plausible reasons: fee capture, sovereignty over state, and the ability to promise its own users a token. A stack that lets you launch with a familiar toolchain and a familiar bridge, under a brand that liquidity providers already recognize, reduces the cost of all three. That is the genuine product.
They are selling shared liquidity. A chain that is part of a recognized cluster gets routed to by aggregators and bridges more readily than a chain that is not. Interop standards are how a cluster turns into a gravitational field.
They are selling regulatory comfort. This is the part that gets written about least and matters most. A stack with a foundation that files documents, a security council with named members, and a public upgrade process is easier to underwrite than a chain with an anonymous multisig. Institutional allocators do not price that in explicitly, but it changes which chains they are allowed to touch at all.
And they are selling a futures contract on influence. Join the club today and you get a seat in the governance that decides what the club becomes. That is a real asset with a real option value, and it is the reason the distribution contest is so intense.
Now look at what is not on that list. Proving system elegance. Sequencer throughput at the ninety-ninth percentile. Daemon-level optimizations. Those things are entry tickets, not differentiators. They get you into the conversation. They do not win it.
I have watched this movie before. During DeFi Summer in 2020, I volunteered inside the Compound governance working group, translating interest-rate models for people who had never opened a bank account abroad. What I saw then โ and what I have seen every cycle since โ is that the protocols that accumulated durable network effects were rarely the ones with the best math. They were the ones whose integration story was easiest to tell to the next person.
That is not an argument for abandoning rigor. It is an argument for putting rigor in the right place. Trust is earned, not mined. On a stack, trust is earned by the quality of the exit you offer when someone wants to leave, and by the clarity of the rules you publish about who can change what.
The Fee Ceiling and the Disappearing Moat
Here is the accounting problem nobody wants to sit with.
A rollup's revenue is sequencer fees plus whatever it can extract from ordering, minus the cost of posting data to Ethereum, minus the cost of proving, minus operations. After Dencun and after the blob capacity increases that followed, the data line item that used to dominate the expense side has collapsed. That sounds like good news. It is, for users. It is much less good news for the rollup, because it also collapsed the scarcity that justified the fee.
Fees on major rollups fell into the fractions of a cent. A chain whose marginal cost of execution approaches zero and whose marginal revenue per transaction approaches zero has a business model problem, regardless of how many transactions it processes. Volume does not save a business whose unit economics are approaching zero from both sides.
So where does the value go? Three places, and each one has a different governance implication.
First, ordering. Whoever decides the sequence of transactions can capture arbitrage, liquidations, and backruns. On most rollups today, that is a single centralized sequencer operated by the founding team. It is fast, it is convenient, and it is exactly the kind of authority that a securities regulator or a market-structure regulator will eventually want to name. A decentralized sequencer set is technically harder, but it is also the only configuration in which the chain's revenue does not look like a private toll road.
Second, native gas tokens and staking. If the chain's token is used to pay fees, to stake with validators, or to buy back and burn against sequencer revenue, then token value becomes a function of activity on that specific chain. This creates an incentive for the stack to maximize the number of chains, because more chains means more total fee flow into a shared token system, even if each chain is individually unprofitable.
Third, interop and bridging fees. As the number of chains grows, the connective tissue becomes the scarce asset. Whoever operates the canonical bridge between cluster members sits on a chokepoint with real pricing power.
The moat is no longer the rollup. The moat is the topology.
This has a consequence that the market has not priced: a stack operator's interests and an individual chain's interests are not the same. The stack wants more chains, because more chains means more aggregate flow and more governance surface. The chain wants more flow to itself. When those diverge โ and they will, the moment two club members compete for the same liquidity โ the governance process becomes a mechanism for adjudicating between members, not a mechanism for serving users.
I have read a lot of governance charters in the last three years. The ones that age best are boring: they contain explicit exit rights, explicit upgrade timelocks, explicit conflict-of-interest rules, and explicit definitions of who is liable when the treasury is drained. The ones that age worst are poetic. They speak in the language of movements and say nothing about who signs the transaction.
Soul in the machine is not a slogan about art. It is the claim that the boring parts โ the timelock, the exit, the disclosure โ are where the ethics actually live.
Who Owns the Block
There is a quiet technical battle underway that will matter more than any proving-system comparison, and almost no retail participant knows it exists. It concerns who gets to say what the next block contains, and how quickly they can say it.
Rollups today achieve fast user experience through a sequencer that orders transactions immediately and posts them to Ethereum later. That ordering promise is a soft commitment. If the sequencer disappears, the chain stalls until someone else takes over โ which, in most designs, means the security council, which means the foundation, which means the entity whose legal status was row thirty-three of that spreadsheet.
The proposed fixes are known: shared sequencer networks, based sequencing in which Ethereum proposers themselves order rollup blocks, and preconfirmation markets in which a party stakes capital against a promise about ordering. Each of these has a distinct trust profile.
Based sequencing is the most philosophically appealing and the most commercially difficult. It inherits Ethereum's liveness and neutral ordering but surrenders the low-latency fee stream that makes a rollup profitable in the first place. Shared sequencers introduce a new committee, which is to say a new cartel, which is to say a new governance question. Preconfirmations are essentially financialized service-level agreements: the guarantee is only as good as the slashable collateral and the dispute-resolution process behind it.
I spent part of 2021 building non-transferable identity credentials with a small collective, five hundred people in a Discord I moderated for six months straight. What I took from that year, and I say this as someone who loves the technology, is that most "decentralization" claims collapse under a simple test: what happens to user funds when the group that runs the system stops answering email? Run that test on the design you are considering. If the answer involves a foundation, a council, or a multisig, you have not removed the trusted party. You have renamed it.
That is why I find the current sequencing debate more consequential than the proof-system debate. Proofs determine how confidently you can verify. Sequencing determines who you have to trust while you are waiting.
And here is the part that connects it all back to row thirty-three. If the sequencer is centralized, the entity operating it is doing something that looks a great deal like operating a market. If it is decentralized through a token, then the token holders โ or their delegates โ are collectively making decisions about order flow, and they are making those decisions without any of the legal architecture that normally accompanies such power. Nobody signed an operating agreement. Nobody filed a registration. The DAO has no legal status, and that absence is not a formality. It is the shape of the risk.
The Two Blank Rows
Let me finally say plainly what I have been circling.
Row thirty-one, who holds the upgrade keys, is the most important question in the stack era, because in a world of federated chains, an upgrade is not a local event. A shared bridge specification or a shared messaging standard means that a change voted through in one governance forum can alter the security assumptions of a dozen sovereign chains that never agreed to it. When I look at how upgrade authority is configured in practice, I look for specific things: the threshold of the multisig, the identities and jurisdictional distribution of the signers, the timelock duration, whether there is any user veto, and whether the code that enforces the timelock is itself upgradeable. I have seen configurations with timelocks that can be reduced by the same multisig that the timelock is meant to constrain. That is not security theater. It is security with a fire exit labeled "authorized personnel only."
Row thirty-three, whether the governing entity has legal personality, is the question that the industry has spent a decade refusing to answer. Most DAOs are unincorporated associations. In the common-law tradition, an unincorporated association that carries on business for profit is frequently treated as a general partnership, which means that individual members can, in the wrong jurisdiction and the wrong fact pattern, be exposed to liabilities far beyond anything they put in. The CFTC's action against Ooki DAO made this concrete: the Commission obtained a default judgment against a decentralized autonomous organization and against its token holders as members. Read the order. Then read your own governance docs. The gap between them is the gap between a community and a defendant.
There are fixes now that did not exist five years ago: the Wyoming DAO LLC, Vermont's blockchain-based LLC, Marshall Islands registration, and the newer unincorporated nonprofit association forms that several law firms have proposed as a model act. Each of them trades something. Wrapping a protocol in a legal entity clarifies liability but introduces a board, and a board is a governance layer that token holders do not control. Not wrapping it preserves the aesthetic and preserves the exposure.
The industry has spent a decade optimizing the part of the system that can be verified and neglecting the part that can be sued. That is not a technology failure. It is a prioritization failure, and the bill arrives during the part of the cycle when everyone has stopped reading carefully.
I will make a prediction that I hope is wrong. Sometime in the next two years, a stack's governance vote will change a shared component in a way that materially damages a chain that voted against it, and the resulting dispute will be resolved not by a forum, not by a vote, but by litigation in a court that has never heard of the protocol. The chain that survives that moment with its treasury intact will be the one that answered row thirty-three before it needed to.
The Pragmatism Test
Now let me argue against myself, because a thesis that cannot survive its own strongest counterargument is not a thesis.
The strongest version of the opposing case runs like this: distribution-driven competition is not a corruption of the technology; it is how technology has always worked. VHS beat Betamax. Windows beat better operating systems. USB-C beat technically superior connectors. Standardization is not a betrayal of merit; it is the mechanism by which merit reaches anyone. Lower switching costs, shared liquidity, and regulatory comfort are not hollow marketing โ they are the things that let a user actually hold an asset without a research department. In that reading, the stack war is a public good, and the fact that the winning stack wins on business development rather than proving elegance is a sign of health, not decay.
I take that argument seriously. It is why I am not writing a piece that says "the better prover should have won." It shouldn't. Engineering elegance has never been a business model.
But the pragmatist case has a blind spot, and the blind spot is the one I keep coming back to. Technical standardization produces compatible products. Governance standardization produces compatible authority. Those are not the same thing. When a phone manufacturer adopts USB-C, it is adopting a connector; it is not adopting a committee that can push an update into phones it does not own. When a chain adopts a stack, it may be adopting both the connector and the committee.
That is the crux of my disagreement with the consensus, and I want to state it as precisely as I can. The consensus treats the stack war as a race to deliver the best product to the most chains. I think the more accurate description is a race to accumulate governance surface โ the right to make decisions that bind parties who are not in the room. Product quality is the bait. Governance surface is the catch.
Look at how security councils are constituted. Look at how upgrade proposals are ratified. Look at how a foundation's board is appointed and by whom. Look at whether a chain that joins the cluster retains the right to fork the shared components, and whether it retains the right to exit the shared bridge without a permissioned process. In most published frameworks, the answers to those questions are either vague or unfavorable to the joiner. The word "sovereignty" appears prominently. The mechanisms that would enforce sovereignty are usually less prominent.
This is where the regulatory dimension becomes unavoidable rather than incidental. I have argued for years that the SEC's approach to digital assets has not been a case of regulators failing to understand the technology. It has been a case of a commission that chose to govern through the sequencing of enforcement actions rather than through published rules, because enforcement preserves maximum discretion. When rules are withheld, they are withheld deliberately. And when rules are withheld, the parties with the best lawyers and the deepest compliance budgets become the ones most able to operate โ which is precisely the opposite of the decentralization story the industry tells about itself.
After the shift in SEC leadership in 2025 and the introduction of more explicit legislative frameworks, the texture of the rules changed but the underlying dynamic did not entirely disappear. Disclosure still matters less than it should. Structure still matters more than it should. The stack that can present a clean corporate diagram to a counterparty's counsel will win accounts that the stack with a cleaner proof system will lose.
So let me run the pragmatism test honestly. If the winning stack is winning because it made institutional adoption easier, and institutional adoption is what brings the next billion dollars of settlement volume onto public infrastructure, then the outcome is good for users even if it is bad for purists. I will grant that. I will grant it fully.
But then apply the test to the exit. If the winning stack's advantage comes from a governance club, what happens to a chain inside that club when the club's interests and the chain's interests diverge? What is the practical remedy? Migration is expensive and slow; forking shared infrastructure is expensive and slow; governance is where the conflict is, so governance is not the remedy. The most likely outcome is that the chain absorbs the cost and the club continues. That is not decentralization failing. That is decentralization never having been implemented in the place where it mattered.
DeFi must mature. The maturation path is not better proofs. It is publishable, enforceable rules about who can change what, with exit rights that a solo developer can exercise on a Sunday afternoon without asking anyone's permission.
I could be wrong about all of this. Here is what would change my mind. I would change my mind if a major stack published an exit specification that let a member chain leave the shared bridge in a bounded number of blocks, unilaterally, with a specified and capped cost โ and then executed that exit at least once, under real conditions, with real funds. I would change my mind if a stack's security council were subject to a genuine user veto with a live threshold, and that veto were invoked and honored. I would change my mind if a significant DAO adopted a legal wrapper that preserved token holder control while eliminating personal liability, and disclosed the result.
None of those things is technically impossible. Each would be a better advertisement for the stack than any throughput benchmark.
What I Watch Instead of Metrics
I want to leave behind a practical instrument, because diagnosis without method is just opinion.
When I assess a stack or a chain now, I run it past a short list, and none of the items on the list appear in most pitch decks.
I ask who can move the state root. That is the question behind all other questions, because whoever can move the state root can move the money.
I ask what the timelock actually enforces, and then I read the contract that enforces the timelock, because a timelock that can be shortened by the same keys is a press release, not a mechanism.
I ask what the exit path costs a user in practice โ not the nominal bridge fee, but the full cost of moving a position out during a period of congestion, including the slippage on the exit and the delay in settlement. Exit cost is the only honest measure of sovereignty I have found.
I ask who is liable when the treasury is drained, and I want the answer in a legal document with a jurisdiction and a date, not in a governance forum post.
I ask how many separate entities control the components that users must trust, because concentration of authority is a risk that hides well across corporate structures.
I ask what happens to a chain that votes against a shared upgrade, and whether the answer is written anywhere.
Six questions. In my experience, a project that answers all six clearly is rare, and a project that answers all six clearly and publicly tends to survive its own success. Trust is earned, not mined. It is earned in documents like those.
The Long View
I am writing this during a bull market, which means most of the people reading it are having a very good quarter and would prefer to think about price. I understand that. I spent a decade learning that asking hard questions during an expansion feels like scolding at a wedding.
But I also spent three months in a New York apartment during the last winter reading forty whitepapers from projects that had died, and I can tell you that the pattern behind their failures was not weak technology. It was unexamined structure. Every one of them had a chain that worked and a governance model that did not, and the market did not discover the difference until it was too late to matter.
The stack era is repeating that pattern at a larger scale, with better marketing and more institutional money behind it. We are building a landscape of hundreds of chains connected by bridges whose upgrade authority is concentrated, governed by assemblies with no legal personality, coordinated by governance processes that were designed to distribute influence and have ended up distributing exposure instead.
Conscience over consensus. The consensus will tell you that this is a race, and that the fastest stack should win, and that the questions I keep asking are the questions of a person who does not understand that growth requires compromise. My conscience tells me that a system which cannot say who is in charge has not distributed power. It has only obscured it, and obscured power is the least accountable kind.
So here is the question I want to leave with you, and I mean it as a builder's question rather than a critic's. If the next wave of the internet is going to be settled on a federation of chains governed by a handful of upgrade keys and an assembly with no legal address, then who exactly did we decentralize it to?
Answer that, and the rest is engineering.