Lab: free tools · size first · stock scanner · crypto scanner · journal

Blockchain Technology

L2 Sequencer Decentralisation: Progress on Arbitrum, Optimism, and Base

Current Ethereum L2 rollups (Arbitrum, Optimism, Base, zkSync) rely on centralised sequencers, single operators that order transactions and batch them to L1. This setup creates censorship and liveness risks. Multiple decentralisation projects are underway: Arbitrum's "BoLD" dispute protocol, Optimism's "Stage 2" roadmap, and shared sequencer networks (Espresso Systems, Astria) aim to eliminate the single sequencer trust assumption while maintaining L2 performance.

The Current State: Centralised Sequencers

Every major Ethereum L2 rollup in production today (Arbitrum, Optimism, Base, zkSync Era, Starknet) relies on a single centralised sequencer operated by the development team. The sequencer is the entity that: receives user transactions, determines their order within blocks, batches them efficiently, and posts them to Ethereum L1 as compressed calldata or blobs. Users interact with the sequencer directly for fast "soft confirmations" (typically 100–500ms) before the transaction data is finalised on Ethereum.

This centralised sequencer architecture creates three categories of risk that are well-understood by the L2 ecosystem and are actively being addressed:

1. Liveness risk: If the sequencer goes offline, the L2 stops processing transactions. Users can still submit transactions directly to L1 through L2 "escape hatches" that bypass the sequencer. But this is slow and expensive. It effectively degrades the L2 to L1 performance during sequencer downtime. Arbitrum and Optimism have both experienced sequencer downtime events that halted the chain for minutes to hours.

2. Censorship risk: A centralised sequencer can choose to exclude (censor) specific transactions. It can refuse to include transactions from specific addresses or protocols. This is analogous to a bank refusing to process payments to a specific merchant. Current L2s provide escape hatches to force-include censored transactions through L1, but the timelock (typically 24 hours for force inclusion) creates a significant window of effective censorship.

3. MEV extraction risk: The sequencer controls transaction ordering, giving it maximum extractable value (MEV) extraction power. It can front-run users, sandwich transactions, or sell priority ordering to sophisticated actors. Centralised sequencers have chosen various policies about MEV. Some share MEV revenue with users through rebates, and some extract privately. Either way, users must trust that the sequencer won't abuse its ordering power, and no cryptographic guarantee backs that trust.

The L2 Stage Framework: Measuring Decentralisation

L2Beat's Stage framework (Stage 0, 1, 2) provides a standardised way to assess how decentralised and trustless an L2 actually is:

Stage 0: The project calls itself a rollup, posts its L2 state roots and data to Ethereum, and publishes software that lets anyone rebuild the L2 state from that data. L2BEAT describes a Stage 0 project as fully controlled by a few entities, so users must trust its operators. Most L2s launched in Stage 0.

Stage 1: Outside of bugs, the only way to block a withdrawal indefinitely or push an invalid L2 to L1 message is to compromise at least 75% of the security council multisig. Upgrades started by anyone outside the security council must give users at least 7 days to exit. Users get real protection, and the security council keeps the power to step in.

Stage 2: The fraud proof system is open to anyone. Users get at least 30 days to exit before an unwanted upgrade, including one a DAO approves, and the security council may act only on errors detected on chain. L2BEAT describes a Stage 2 project as fully controlled by code, so security rests on the proof system and Ethereum's consensus. This is the end goal for serious L2s.

As of 2026, most major L2s are at Stage 0 or Stage 1. Achieving Stage 2 while maintaining performance and supporting the whole range of EVM functionality is technically challenging. It requires strong dispute resolution systems that can handle complex contract interactions without false-positive fraud proofs.

Arbitrum BoLD: Decentralised Dispute Resolution

Arbitrum's "BoLD" (Bounded Liquidity Delay) protocol is a major upgrade to Arbitrum's fraud proof system designed to enable permissionless, decentralised dispute resolution. The main innovation: BoLD bounds the maximum dispute resolution time regardless of how many challenges are made. That prevents adversaries from using repeated challenges to indefinitely delay L1 settlements. Such delays were the vulnerability in Arbitrum's original challenge protocol.

BoLD transitions the security model from "trusted validator set" to "permissionless validation". Under the old model, only Arbitrum-approved validators could challenge state roots. Under the new one, anyone who stakes ARB tokens can participate in state root challenges. This is a prerequisite for Arbitrum achieving Stage 2 status. Arbitrum's phased rollout of BoLD has been underway through 2025–2026, with full deployment expected to coincide with the removal of the "training wheels" security council override capability.

Optimism Superchain and Stage 2 Roadmap

Optimism's roadmap focuses on the "Superchain". That is a shared infrastructure layer where multiple OP Stack chains (Optimism, Base, Mode, Zora, and many others) share sequencing, dispute resolution, and security infrastructure. The goal: decentralised sequencing across the entire Superchain network, where validator sets are shared and no single team controls any individual chain's sequencer.

Optimism's "Stage 2" roadmap includes a decentralised fault proof system, governance minimisation, and eventually a shared sequencer set for the whole Superchain. Fault proofs are live on Optimism as of 2024, but they still carry security council override capability. That puts Optimism at Stage 1 currently. Governance minimisation means reducing the Optimism Foundation's direct control over protocol upgrades. Base (built by Coinbase on OP Stack) inherits whatever decentralisation properties OP Stack provides. That means Base's sequencer is currently Coinbase-operated, and decentralisation awaits Superchain sequencer sharing.

Shared Sequencer Networks: The Horizontal Solution

An alternative to each L2 developing its own decentralised sequencer is shared sequencer networks. These are third-party sequencing services that multiple L2s can use simultaneously. Key projects:

Espresso Systems: Building a decentralised sequencer network using HotShot consensus (a variant of proof-of-stake BFT consensus) that can provide fast, censorship-resistant sequencing as a service to any L2 that integrates it. Espresso's architecture promises sub-second soft finality with decentralised validation. That would address both the centralisation risk and the latency of waiting for L1 finality.

Astria: A shared sequencer specifically designed for rollups, providing a shared mempool and decentralised ordering layer that rollups can post to, while retaining their own execution environments and L1 settlement.

Shared sequencers also enable atomic cross-rollup composability: the ability to execute a transaction that spans multiple L2s in a single operation. One example is a swap on Arbitrum followed immediately by a deposit to Optimism. It works because both chains share the same sequencing layer and can commit to including both sides atomically.

Summary

L2 sequencer decentralisation is one of the most important ongoing development priorities in the Ethereum ecosystem. It addresses the gap between the theoretical security guarantees of rollups (inheriting Ethereum's security) and the practical trust requirements of current centralised sequencer designs. Arbitrum BoLD, Optimism's Stage 2 roadmap, and shared sequencer networks like Espresso represent the active technical approaches. For users and developers evaluating L2s, the L2Beat Stage framework provides the most clear-eyed assessment of how much trust each L2 actually requires today versus its fully decentralised end state. That is a critical distinction for applications requiring censorship resistance or users moving substantial capital.