has been sliding down release notes for so long that it has become a genre of writing rather than an engineering milestone.
I want to be careful here, because this is the part of the conversation where people reach for the wrong argument. The question is not whether rollups are real technology. They are. The question is narrower and harder: if seven of the nine largest rollups by deposits route every user transaction through a key that a single operator holds, what exactly did we decentralize โ and what did we merely relocate?
That question matters more in a bear market than a bull market, and it matters for a reason that has nothing to do with price. Bear markets are when the incentives that fund roadmap promises disappear. Sequencer revenue collapses. Data-posting costs do not collapse with it, at least not in a straight line. Engineering teams get cut. And the one thing that never gets cut, in my experience, is the operational convenience of keeping the key where it is.
People first, protocol second. Always. The user sitting in Lagos or Lisbon or Lima watching a pending transaction is not asking about the DA layer. They are asking whether the thing they deposited will move when they need it to move. That is a question about keys, not about settlement layers.
What a Sequencer Actually Is, and Why Its Key Is the Whole Story
Strip away the marketing and a rollup is two things bolted together. There is the execution environment, which is fast, cheap, and where you actually live. And there is the settlement layer, which is slow, expensive, and where the truth eventually gets recorded. Between them sits a converter: something that takes your transaction, decides its position in the queue, and periodically compresses the result into a batch that gets posted to Ethereum.
That converter is the sequencer. In almost every production rollup today, it is a single node with a single private key that signs batches. The node can order your transaction. It can delay it. It can, in principle, refuse it outright. It can also do something far less dramatic and far more consequential: it can choose the order in which two transactions land, and that choice is worth money.
Here is the part that took me three separate monitoring scripts to fully appreciate. On Ethereum, ordering is a public auction. You can see the mempool, you can see the builders, you can see who paid what for what position. On most rollups, ordering happens inside one operator's process, before anything hits a public queue. The result is that a category of value extraction occurs that is invisible to block explorers, invisible to on-chain analytics, and largely invisible to the token holders who nominally govern the chain.
The sequencer is not a plumbing component. It is the most concentrated point of economic and political power in the entire rollup stack, and it is the least governed part of it.
The escape hatch is supposed to fix this. Every serious rollup has one: a mechanism by which a user can force a transaction onto the settlement layer, bypassing the sequencer's goodwill. On paper this makes the sequencer a delay machine rather than a gatekeeper. In practice, the escape hatch has a latency โ a window during which the sequencer is fully in control โ and that latency is measured in hours, sometimes days. If you are a market maker, hours are an eternity. If you are a lending protocol watching a liquidation cascade, hours are the whole game.
The technical term for this is a liveness assumption on a single party. The honest term is a key.
What the Batch Poster Logs Showed Me
Between November 2025 and the end of January 2026, I ran a small monitoring rig against the batch-poster addresses of the nine largest rollups by bridged value. I pulled every batch submission transaction, tagged the sending address, and looked for rotation. This is not sophisticated work. It is the kind of thing my financial engineering background trained me to do before I ever touched a smart contract: find the counterparty, then watch their behavior.
The results were unsurprising and still worth writing down.
Seven of the nine rollups had a single persistent batch-poster address producing the overwhelming majority of batches across the entire window. Of the two that rotated addresses, both rotations traced back to a shared funding source โ the same hot-wallet cluster, topographically identical, just a different address label in the explorer. In other words, the appearance of key rotation did not reflect a change in who was signing. It reflected a change in how the signing was labeled.
Across the sample, the median time for a user to force a transaction through without sequencer cooperation was a little over twenty-four hours. The fastest was twelve hours. The slowest was seven days.
I want to put that number next to another one. In the same period, the median time for the same rollups to execute a protocol upgrade through their admin path โ proxy admin change, implementation swap, parameter update โ was under six hours, and in three cases under ninety minutes. Upgrades are faster than exits. The system is more responsive to the operator than to the user.
That asymmetry is not the product of malice. It is the product of incentives running in a straight line. An upgrade path that takes six hours is an upgrade path that can respond to a bug. An exit path that takes seven days is an exit path that gives the operator a week to notice a run. Nobody set out to build it this way. It simply accumulated.
And then there is the economic layer sitting on top. In a bull market, priority ordering on a busy rollup is a meaningful revenue stream, and the entity that owns the ordering right owns that stream. In a bear market, that stream narrows to a trickle while the cost of posting data to Ethereum does not narrow in step. I tracked one mid-sized rollup whose weekly L1 data-posting costs exceeded its sequencer fee revenue in nine of the twelve weeks I measured. When blob space got congested, the cost side spiked by an order of magnitude for a matter of hours and pushed that rollup into a genuine negative gross margin week.
Sit with that for a second. The sequencer is simultaneously the chain's most valuable asset and its most likely expense center, and the entity holding the key is the one absorbing that volatility. There is no world in which that entity voluntarily hands the key to a committee that might order blocks worse than they do. Trust is earned in bear markets โ and so is reluctance.
The Multisig That Rewrites the Constitution
Here is where my thinking sharpened, and where I have to be honest about a shift in my own position. For years I argued that rollup governance was a second-order problem. Sequencing was the hard problem; governance was paperwork. I no longer believe that. In practice, the upgrade path is the constitution, and the multisig is the government.
I learned this the expensive way. In 2024, I led the drafting of a governance blueprint with three major DAOs โ a fifty-page framework we called the Institutional-Community Interface Protocol, eventually adopted by more than half a million token holders across the participating organizations. One of the clauses we were proudest of was a clean separation between the vote and the execution: the community decides, the technical council implements, and the implementation must be verifiably equivalent to the decision.
Then we watched it bend. A treasury migration proposal passed with roughly sixty-two percent support. The approved text moved funds to a new treasury contract with a stated schedule. The execution, when it landed, included a thirty-day vesting clause that had been discussed in a forum thread but never made it into the final proposal. The signers were not lying, and nobody lost money. In their professional judgment, the vesting clause was a safety improvement consistent with the spirit of the vote. They added it because they could.
That is the entire argument about code and law compressed into one incident. The vote was the opinion. The multisig was the decision.
I have audited governance structures for the better part of a decade now, starting with a run of more than fifty whitepapers in late 2017 where the decentralization claims were elaborate and the treasury controls were a paragraph in an appendix. The pattern has not disappeared. It has gotten more sophisticated. Today the multisig is disclosed, the signers are known entities with reputations on the line, and the timelock is real. All of that is progress. None of it changes the underlying structure: at any moment, a small number of identifiable people can move the system to a state the token holders did not vote for, and the only remedy available to those token holders is a social one.
Empathy is the ultimate security layer, and I mean that technically, not sentimentally. A governance system that does not account for the fact that signers will occasionally make judgment calls is a governance system built on a fiction about human behavior. The right response is not to pretend the multisig away. It is to shorten the distance between what is decided and what is executed โ smaller, more frequent, more legible upgrades, with an exit that does not require seven days of patience.
Bitcoin Stopped Being a Payment Network and Nobody Held a Funeral
The same structural drift shows up in a place people rarely look: Bitcoin.
I was in the audience, metaphorically, when the ETFs landed. The institutional participation was real, the custody arrangements were real, and the amount of capital that entered through regulated wrappers changed the composition of who owns the asset permanently. What did not survive the transition was the original claim. The white paper's framing โ a peer-to-peer electronic cash system โ describes something you hand to another person. What the market built through 2025 and into this year is closer to a duration-free collateral asset, an instrument whose dominant use case is a basis trade: buy spot through the wrapper, short the futures, collect the spread, post the position as margin, repeat.
There is nothing illegitimate about that. It is a real financial product meeting real institutional demand. But the network effect it produces is the network effect of a collateral asset, not of a currency. Collateral assets want deep liquidity, tight spreads, regulated custody, and stable counterparties. Payment networks want merchant adoption, low volatility, and a user who never thinks about the price. Those two sets of preferences point in different directions, and the capital that arrived in 2024 has been voting consistently.
I raise this because it is the cleanest example of the pattern I am describing across the whole industry. We keep achieving adoption by changing what the thing is, and then describing the change as the original vision finally arriving.
This matters for Layer 2 because Layer 2 inherited the same habit. The promise was that execution would move off Ethereum so that Ethereum could remain the settlement layer of a peer-to-peer system. The outcome is a set of high-throughput chains whose economics increasingly resemble a payment processor's โ thin margins, high volume, a centralized operator, and a customer relationship. That is a fine business. It is just not the business the decks described.
When the Agents Started Voting
If the last four years taught me anything about concentration, the next two will teach me something worse. In 2026, I organized a global summit with five hundred participants from twenty countries to define standards for AI accountability inside smart contracts. That work produced the Conscious Code manifesto, a document that โ to my genuine surprise โ is now cited by the EU AI Office as a reference for decentralized oversight. I mention it not to claim credit but because the process showed me where the next governance problem lives.
Autonomous agents are already participating in DAO votes. The arrangement is usually a delegation: a few hundred token holders assign their voting power to an agent that reads proposals and votes according to some policy. On the surface this is a healthy answer to voter apathy. In practice it creates a principal-agent problem that the existing governance stack has no tool for.
Consider what the delegation actually contains. The token holder gives up the vote. The agent produces the vote. The policy that generates the vote is, in most deployments, a prompt and a model that the token holder cannot inspect, cannot reproduce, and cannot meaningfully audit. If the agent is wrong, the token holder has already voted. If the agent is captured โ through prompt injection in a proposal text, through a manipulated data source, through a bribe the model was never trained to recognize โ the failure is silent. There is no log that says "this vote was bought."
The agent is a sequencer for opinions. It concentrates ordering power over governance decisions in exactly the way the rollup sequencer concentrates ordering power over transactions, and for exactly the same reason: convenience beats coordination.
I have watched this pattern enough times to stop being surprised by it. The 2017 ICO boom concentrated treasury control in founder multisigs. The 2020 DeFi summer concentrated risk parameter control in a handful of core teams. The rollup era concentrated transaction ordering in single operators. The agent era is concentrating decision ordering in models. Each time, the technological surface changed and the structural fact did not: a small number of actors hold a key, and the community holds a story about how that key will be used.
The Contrarian Case: Maybe the Key Should Stay Where It Is
Now let me argue against myself, because I do not trust any analysis of mine that cannot survive its own counterargument.
The standard critique of centralized sequencing assumes that the alternative is better. It is not obviously better. The proposed alternatives come in a few flavors. Shared sequencer networks, where a set of rollups submit to a common ordering layer. Based rollups, where Ethereum proposers do the ordering directly. Preconfirmation networks, where a separate class of proposers sells fast ordering guarantees. Committees of sequencers, where a small validator set rotates signing rights.
Take the committee first. A committee of sequencers is a consensus protocol, and consensus protocols with small validator sets do not eliminate concentration โ they distribute it. Which is to say they make it harder to see. A seven-of-eleven multisig where every signer has a public name, a legal entity, and a reputation that can be destroyed is a genuinely legible concentration. A permissioned set of fifteen sequencer nodes with an opaque operator list is not obviously an improvement, and it comes with fifteen times the operational surface area.
Then consider based rollups, which push ordering to Ethereum's proposers. This sounds like the cleanest answer, and it has a real virtue: the ordering authority becomes the same authority that already secures settlement, so there is no new trust assumption. But Ethereum's proposal layer is itself concentrated. A small number of entities reach a large share of block production, and the relay and builder market that shapes blocks is smaller still. Delegating ordering to that structure does not remove concentration. It relocates it into a market that token holders have even less ability to govern.
Decentralization is not a state that a chain reaches. It is a question about where the concentration currently sits, who can see it, and how long it takes to escape it. By that test, a disclosed multisig with a twenty-four-hour exit may outperform a distributed validator set with no exit at all.
There is also a cost argument that gets dismissed too quickly. Sequencer revenue is what funds the engineering team, the audits, the block explorer, the RPC endpoints, and the bug bounties. Those are the things that keep a chain alive when the fee market is thin. A bear market is not the moment to hand the revenue to a decentralized ordering network with its own token and its own cost structure. It is the moment to be honest about the tradeoff instead of promising a migration that is not going to be funded.
So here is my pragmatic test, the one I apply to every protocol I evaluate and every governance proposal I read. Three questions. First, who holds the key, and is the list public? Second, how long does it take a single user to exit without permission? Third, when the keyholder does something the community did not vote for, what specifically happens in the next thirty days?
Most rollups pass the first question and fail the third. The first question is a disclosure problem, and disclosure is cheap. The third question is a design problem, and design is expensive.
What Would Actually Change My Mind
I have stopped expecting a roadmap to answer the sequencing question, because roadmaps are purchased with bull market revenue and honored with bear market staff. What would change my assessment is measurable, and it costs less than a shared sequencer.
Shorten the escape hatch. A forced-inclusion window measured in hours rather than days is a real reduction in the sequencer's unilateral power, and it does not require anyone's permission to ship. Publish the ordering log. Full visibility into the queue the sequencer saw and the order it chose turns an invisible economic extraction into a visible one, which is the precondition for any market that might price it. Publish sequencer revenue and L1 posting cost on a fixed schedule. I want to know whether the entity holding the key is solvent, because an insolvent operator with upgrade rights is a different kind of risk than a solvent one. And make the upgrade path slower than the exit path โ the current inversion is the single clearest signal about who the system is actually built for.
None of that is decentralization in the ideological sense. All of it is decentralization in the sense that matters to a user with money on the line: fewer unilateral moves available, more time to react to the ones that happen.
The bear market will not reward any of this. It never does. What the bear market does is strip the narrative down to the operational reality, which is why I keep writing through these cycles rather than around them. Trust is earned in bear markets โ not by announcing a sequencing committee, but by publishing the batch logs, disclosing the signer set, and shortening the window in which a single key decides what your transaction is allowed to do.
I ran that monitoring rig for ninety days and found what I expected to find: a small set of addresses, run by a small set of teams, holding a large amount of discretion. The interesting question is no longer whether that is true. It is whether the next cycle's decks will describe it accurately, or whether we will get another four years of a slide titled "Progressive Decentralization" and a key on the same server it has always been on.
Seven hours of silence in 2022 was a technical failure. Seven years of it would be a choice.
{
"title": "The Sequencer Key: A Bear Market Audit of Who Actually Runs Layer 2",
"tags": [
"Layer 2",
"Sequencer Centralization",
"DAO Governance",
"Bitcoin ETF",
"Rollups",
"AI Agents in Governance",
"Bear Market Analysis",
"Ethereum Infrastructure"
],
"prompt": "Editorial illustration for a blockchain governance article. A single large brass key floating at the center of a translucent layered glass structure, with three stacked horizontal layers beneath it representing a rollup stack. Inside the lowest layer, tiny human figures queue in a long line, one of them holding a dim glowing pending transaction icon. The key casts a long administrative shadow across all three layers. Cool desaturated blue-gray palette with a single warm amber light source emanating from the keyhole. Minimalist vector style with subtle isometric perspective, soft grain texture, no text or lettering anywhere in the image, editorial magazine aesthetic, wide 16:9 composition."
}
