On the morning of 22 March, a five-line Python loop I keep running against four of the largest rollups returned something I had not logged since the last bear market. One chain reported a null block-production interval for 41 consecutive minutes. Then it resumed at the identical gas price it had abandoned, as if the gap had never happened. No failover. No emergency governance call. No public post-mortem for six days.
Three of those four rollups were being ordered by one sequencer. One process — or one tightly coupled cluster — deciding the position of every transaction in every block. Nine months into tracking this the way I once tracked ETF net inflows, the structural number had not moved: the count of production rollups with a genuinely decentralized sequencer set — more than one independent operator, carrying a credible liveness and safety guarantee — remains approximately zero.
That is the anomaly. Not the outage. The fact that a bull market, a token-unlock cycle, and forty-plus L2 networks have not budged it.
For anyone who has been in the space but not inside the architecture: an L2 sequencer accepts user transactions, orders them, executes them against the rollup state, and periodically publishes compressed batches — plus, depending on the design, proofs — to Ethereum L1. Optimistic rollups (Arbitrum One, OP Mainnet, Base) post fraud proofs against a challenge window. ZK rollups (zkSync Era, Starknet, Scroll) post validity proofs. Either way, the sequencer is the ordering authority, and ordering is where value lives.
Today, nearly every production sequencer is operated by a single entity. Arbitrum's runs under Offchain Labs. OP Mainnet's runs under OP Labs. Base's runs under Coinbase. The 'decentralized sequencing' category — shared sequencers such as Espresso, Astria, and Flashbots' SUAVE; based rollups that inherit ordering from L1 proposers, such as Taiko; and rotating-operator sets — has shipped testnets, grant programs, and conference talks. It has not shipped, at scale, into a live chain carrying meaningful value at risk.
The cleanest public scorecard is the L2Beat 'Stage' framework. Stage 0 is full training wheels: a security council can override the system and the operator set is centralized. Stage 1 is limited training wheels, with a functioning proof system and a governance mechanism that cannot unilaterally upgrade contracts. Stage 2 removes the training wheels entirely. As of this writing, the overwhelming majority of live rollups sit at Stage 0 or Stage 1. The high-throughput, high-TVL chains — the ones whose tokens trade at the richest multiples — are not the ones closest to Stage 2.
I have argued before that 'decentralized sequencing' has been a PowerPoint for two years. Here is the evidence chain that produced the claim, and why I think the market is still misreading it.
Three data streams matter: liveness, revenue, and upgradeability. Liveness first. Sequencer outages are not hypothetical: Arbitrum's sequencer halted during a routine software upgrade in early 2023, and OP Mainnet's went offline for roughly two hours in 2022. Each time, the escape hatch — the ability to force a transaction through L1, bypassing the sequencer — existed in the documentation and was, in practice, slow, manual, and rarely exercised. When a rollup tells you it has a censorship-resistant force-inclusion mechanism, ask the next question: what is the latency, in hours, from the user's seat? On the busiest chains, force-inclusion windows run from hours to a day. On a liquidation cascade, a day is forever.
Revenue second. When EIP-4844 shipped in March 2024, blob space collapsed the L1 data cost of posting rollup batches. Fees paid by L2 users fell by roughly an order of magnitude on the busiest chains. But sequencer economics are fees collected minus L1 posting cost — a spread. When the input cost falls and the user fee falls, the operator captures the difference. There is no competition for ordering, so there is no competition on price. That is not a market. That is a margin.
Upgradeability third. A Stage 0 rollup's contracts can be upgraded by a multisig that swaps out the proof system. The threshold matters; the signer identities matter. But the structure is identical: a small set of humans can rewrite the security guarantees of a chain holding billions. The bull market prices throughput. It does not price the fact that the security model is a multisig with a documented owner list.
I came to this frame the hard way. In 2017 I audited a lending contract before its mainnet launch and found the critical defect in the withdrawal logic — not in the math, in the control flow. Sequencer centralization is the same species of flaw. The cryptography is fine. The control flow has one owner. Three years later I built a Python arbitrage bot for Uniswap V2 and Curve Finance, exploiting a roughly $30 spread between DAI on the two venues, executing about 150 trades a day at 99.8% fill accuracy for three months. The lesson was never the profit. It was that in a deterministic system, only two variables matter: ordering and latency. On an L2, both are the decision of a single actor. Everything else — the marketing, the TVL curve, the modular stack diagram — is downstream of that.
Now run the logic forward. Shared sequencers, based rollups, and operator rotation all address ordering and latency. So why has nothing scaled? Because centralized sequencing is a feature of the current growth phase, not a bug. It is faster, cheaper, and simpler to run one node than to coordinate a set. The decentralization is deferred because deferring it is profitable, and because the user experience of a single sequencer is genuinely better. That is a rational business decision. It is not a decentralization roadmap, and it should not be priced as one.
Consider what the governance tokens actually do. ARB and OP confer voting rights over a treasury and, in theory, over protocol parameters. They confer no control whatsoever over the sequencer operator's transaction ordering. You can pass a governance vote and the sequencer can ignore it. You can hold thirty percent of the supply and you still cannot censor — or un-censor — a single block. The token is a claim on governance theatre, not on ordering.
Then there is MEV. On Ethereum L1, proposer-builder separation split block building from block proposing and created a competitive market for order flow. On an L2, the sequencer is both builder and proposer, and it keeps the ordering surplus. Arbitrum's proposed Timeboost mechanism — an auction for an express lane — is an attempt to formalize and sell priority ordering. Read that carefully. The mechanism does not decentralize ordering; it monetizes it, with auction proceeds flowing to the operator and the DAO. The sequencer keeps the steering wheel and starts charging for the passenger seat.
Here is where the consensus is looking the wrong way. The market's stated worry is liveness — what if the sequencer goes down? Liveness risk is at least partially priced into fee discounts, insurance products, and TVL haircuts. The unpriced risk is censorship and legal coercion.
A single sequencer operator is a legally identifiable entity. After OFAC sanctioned Tornado Cash in August 2022 — and after developers were subsequently charged — the compliance surface for any US-linked operator hardened into a chokepoint. Coinbase runs Base's sequencer. Offchain Labs runs Arbitrum's. If a regulator says 'filter these addresses,' the sequencer is one knob, and the operator is one phone call away. The force-inclusion fire exit still exists, but it takes a day to open, and it logs your name.
Second, the causation error. When L2 fees collapsed after March 2024, a great deal of commentary credited competition between rollups. Wrong variable. The fee collapse was blob supply — a change in the input cost of posting data. Competitive intensity was flat. Mistaking a supply shock for a competitive market is the oldest error in the book, and it is currently embedded in a dozen L2 investment theses. When a rollup advertises decentralized sequencing while the sequencer is one machine, treat it as you would any other too-good-to-be-true claim: ask for the operator list, in writing, alongside the failover plan.
Watch two things next quarter. First, L2Beat stage changes: if a top-five rollup by TVL moves to Stage 2, the decentralization premium is real; if none do, it is a fiction. Second, the first based or shared sequencer in production carrying more than a billion dollars in TVL. Until one of those two events prints, the correct posture is skepticism toward the roadmap narrative and hard respect for the margin being extracted in the meantime. The data has not changed in nine months. Ask yourself why the story has.
