cryptonist
Fibre optic strands gathered into a bundle, their ends lit

DeFi

Layer 2 explained: what a rollup moves off Ethereum

A layer 2 rollup executes transactions away from Ethereum’s main chain and posts two things back to it: the data needed to rebuild its state, and a claim about what that state now is. Rollups differ in how the claim is checked, in who puts transactions in order, and in who can still change the contracts that hold users’ deposits. Those three questions separate one rollup from another more than speed or fees do.

What moves off the chain, and what stays on it

Execution and storage are what leave. A rollup runs its own virtual machine, usually a close copy of Ethereum’s, and keeps its own record of balances and contract state. What stays on Ethereum, in ethereum.org’s account of optimistic rollups, is a set of contracts: one that stores the rollup’s batches and its state commitments, and a bridge that holds what users deposited.

Ethereum then does three jobs the rollup cannot do for itself. It makes the rollup’s data available, so anyone can recompute the state. It settles disputes or checks proofs, so an invalid state cannot be finalised. And it offers a way around the rollup’s operator, so a user who is being ignored can still get a transaction in. Ethereum.org’s scaling overview keeps the name layer 2 for designs that take their security from Ethereum in this way. A sidechain, by contrast, is a separate blockchain with its own consensus, joined to Ethereum by a two-way bridge, and a validium, as the zero-knowledge rollups page describes it, verifies on Ethereum but keeps its data somewhere else.

A deposit shows how the parts fit together. The user sends ether or a token to the rollup’s bridge contract on Ethereum, the contract relays the deposit, and an equivalent amount is minted at the address the user chose on the rollup. Whatever backs tokens bridged this way therefore sits in that contract on Ethereum. DefiLlama files such contracts in a canonical bridge category, which as of 22 November 2026, per DefiLlama, shows the Arbitrum Bridge at $2.8B, the Base Bridge at $2.6B and the Optimism Bridge at $750.4M. Under DefiLlama’s rules a bridge’s balance counts towards no chain’s total, one of several reasons what TVL measures gives for not adding figures together across layers.

On an EVM rollup the user’s Ethereum address exists too, controlled by the same key, which is why a wallet that works on one chain and not another treats Ethereum and its rollups as a single family. What Ethereum is covers the base layer itself.

Optimistic rollups: publish first, allow a challenge

An optimistic rollup posts its batches and a new state root without proving that the root is correct. It is optimistic in the literal sense: the contract on Ethereum assumes the claim is valid unless someone shows otherwise within a set period, the challenge period. A challenger who disagrees computes a fraud proof. In the multi-round version ethereum.org describes, the two sides split the disputed computation in half again and again until they disagree about a single step, which the contract on Ethereum then executes itself to decide who is lying. If the challenge succeeds, the operator’s bond is slashed.

Two consequences follow from that design. The first is that the chain’s correctness rests on the existence of one honest party who is watching and willing to challenge, and that party can only build a proof if the data was published. The second is the wait. A withdrawal from the rollup to Ethereum cannot complete until the challenge period has run out, because until then the state it relies on could still be overturned. Arbitrum’s documentation gives the period for a message from Arbitrum to Ethereum as 6.4 days. The documentation for the OP Stack, the software OP Mainnet runs on, says withdrawals through its standard bridge require a minimum seven-day wait.

The same page heads off a common misreading. The delay applies to withdrawals, not to using the chain: an OP Stack transaction becomes final once its data is in a finalised Ethereum block, typically about 15 to 30 minutes after submission on OP Mainnet, and a fault-proof challenge removes a bad withdrawal claim without reorganising the chain.

There is a way to skip the wait, and it changes who is trusted. Ethereum.org describes liquidity providers who take over a pending withdrawal and pay the user on Ethereum straight away, for a fee, having checked the withdrawal themselves. That replaces waiting with a counterparty. Third-party bridges that offer fast exits are separate systems with their own contracts and their own trust assumptions, and the bridge and cross-chain incidents in DefiLlama’s hacks record show what that can add; cross-chain bridge hacks explains how those failures happened.

ZK rollups: prove first

A zero-knowledge rollup reverses the order. Its operator submits a new state root together with a validity proof, a short cryptographic proof that the root is the correct result of executing the batch, and the contract on Ethereum accepts the root only if the proof verifies. Nothing has to wait for a challenge, so ethereum.org notes that an exit executes as soon as the proof for its batch is verified.

The costs sit elsewhere. Producing proofs is computationally heavy, and ethereum.org warns that the specialised hardware it needs may concentrate control in a few parties. Verifying a proof on Ethereum costs gas, which ethereum.org puts at around 500,000 per proof. Some proof systems (ZK-SNARKs) need a trusted setup, which stays secure as long as at least one participant destroyed their share of the secret randomness.

The name misleads in one respect. The same page is explicit that using validity proofs for scaling does not make a rollup’s transactions confidential: the data is published so that anyone can reconstruct the state. A ZK rollup still has to publish it, because without the state data users cannot see their balances or build a withdrawal. Some post compressed transactions and others post only the differences in state. The research group ETHBEAT, formerly called L2BEAT, whose project pages were read for this piece on 22 November 2026, lists ZKsync Era and Starknet as posting state differences and Linea as one of the exceptions that posts transaction data.

The sequencer, and what central ordering means for a user

Ordering usually sits with one party. Ethereum.org says most ZK rollups use a single operator to execute transactions and submit blocks, and describes the sequencer some optimistic rollups use as the entity with priority access to the rollup, the only one authorised to submit batches to the contract on Ethereum, while transactions from anyone else wait in a separate inbox until the sequencer includes them. On Arbitrum, the confirmation a user sees within a second of sending a transaction is what the Arbitrum documentation calls soft finality, a provisional confirmation that depends on the sequencer behaving honestly.

A sequencer can choose the order of transactions, delay them, refuse them or stop altogether. It cannot make an invalid state final, because the proof system stands between its batches and the bridge. The power to order is still worth money, and what MEV is explains how. Centralised sequencing, for a user, therefore means one operator’s ordering on an ordinary day, and on a bad day a question about what the escape route is and how long it takes.

Rollups answer that question differently:

  • Arbitrum One. A user can send a forced transaction to a delayed inbox on Ethereum. If the sequencer ignores it for the force-inclusion window, at most 7,200 Ethereum blocks, which at 12 seconds a block is 24 hours, anyone can move it into the chain without the sequencer. The same page notes that the chain owner can switch on an allowlist for that inbox or pause it.
  • OP Mainnet. On OP Stack chains, transactions sent to the portal contract on Ethereum are processed within the sequencer window, set to 12 hours by default, though it may differ between chains. ETHBEAT gives the same 12-hour maximum delay for forcing a transaction onto OP Mainnet.
  • Base. ETHBEAT lists the same route through Ethereum for Base, with a maximum delay of 12 hours.
  • Some ZK rollups. ETHBEAT’s summary for ZKsync Era says users can queue a transaction on Ethereum but cannot force it, and the sequencers can stop processing the queue for everyone. For Linea it lists no mechanism at all until six months without finalised blocks, after which the operator role becomes public.

Data availability, and what blobs changed

Publishing the data is what lets anyone check the rollup without trusting its operator. Ethereum.org’s data availability page puts it plainly for both designs: without the data, challengers cannot construct fraud proofs against an optimistic rollup, and users of a ZK rollup cannot know their balances. It also separates availability, the assurance that the data could be downloaded when the block was checked, from retrievability, the ability to fetch old data later, which it says the core protocol is not primarily concerned with.

Rollups first posted their batches as calldata, which stays in Ethereum’s history permanently. EIP-4844, which went live in the Dencun upgrade in March 2024, added a cheaper container built for the purpose: the blob.

A blob, as the EIP defines it, is a fixed block of 4,096 field elements of 32 bytes each, 128 KiB, carried in a new type of transaction. The EVM cannot read it. Contracts see only a commitment to the blob, which is enough for a fraud proof or a ZK rollup’s proof to check data against. Blobs are priced in their own kind of gas, whose base fee rises when blocks carry more blob data than a target and falls when they carry less, much as EIP-1559 does for ordinary gas, and the fee is burned.

Nor are blobs kept for ever. The data lives with the consensus layer, whose specification sets 4,096 epochs, about 18 days, as the minimum period over which a node must serve blob data to its peers. Ethereum.org’s Dencun page says blobs are guaranteed to be available for that window and pruned from the network after it. The EIP’s reasoning is that a rollup needs its data to be available long enough for honest parties to reconstruct the state, not permanently.

Capacity has been raised since. EIP-4844 set a target of three blobs per block and a maximum of six. EIP-7691, part of the Pectra upgrade in May 2025, raised them to six and nine. EIP-7892 then created forks that change only blob parameters, and the consensus specification’s mainnet configuration schedules a maximum of 15 blobs from 9 December 2025 and 21 from 7 January 2026. The Fusaka upgrade of 3 December 2025, listed in EIP-7607, included PeerDAS, under which nodes sample blob data instead of downloading all of it.

For a user, the point is that a rollup fee contains a charge for posting data to Ethereum, priced in a market separate from Ethereum’s ordinary gas, which what gas is covers. A chain that keeps its data somewhere else, with a committee of trusted parties attesting to it, is relying on those parties and not on Ethereum for this guarantee.

How much a rollup still depends on its operators

A rollup’s contracts on Ethereum can usually be upgraded, and whoever holds that power, a team, a multisig or a council, can change the rules users deposited under. ETHBEAT’s stages framework grades how far that power has been given up. It is a secondary source, the work of an organisation that describes itself as a public goods company, and its own summary of the idea is that Stage 0 is a project fully controlled by a few entities, Stage 2 one fully controlled by code, and Stage 1 something in between.

The requirements are specific:

  • Stage 0. The project calls itself a rollup, posts state roots and all the data needed to rebuild its state on Ethereum, makes software that can reconstruct that state available, and uses a proper proof system. A fraud proof system must let at least five external actors submit proofs.
  • Stage 1. Apart from bugs, the only way to block a withdrawal indefinitely, or to push through an invalid one, is to compromise at least 75% of a Security Council. Upgrades initiated by anyone outside the council must give users at least seven days to exit.
  • Stage 2. Fraud proofs are open to anyone, users get at least 30 days to exit before an unwanted upgrade, and the Security Council may act only on bugs that can be shown on-chain.

The framework measures decentralisation, not safety. ETHBEAT says so on the framework page and again on its project pages, where a note reads that stages do not reflect project security; a team could deploy a Stage 2 rollup with an experimental proof system or unaudited contracts.

Read on 22 November 2026, ETHBEAT rates Arbitrum One, Base and OP Mainnet at Stage 1, along with Starknet. It rates ZKsync Era and Linea at Stage 0. For Base and OP Mainnet it lists the exit window as none, because the contracts can be upgraded instantly. For Arbitrum One it lists none on the emergency path, on which 9 of the 12 members of the Security Council can upgrade at once, and a window of more than a week only for ordinary upgrades. For Base it notes that an upgrade needs the approval of both the Base Coordinator Multisig and the Base Security Council. On the OP Stack side, the fault-proof explainer describes the backstop in its own words: the Optimism Security Council, acting as the Guardian, can pause withdrawals, blacklist dispute games or fall back to a permissioned system.

So Stage 1 does not mean nobody can change the rules. It means that blocking or forging withdrawals, bugs aside, takes a council supermajority.

The figures, and what each one counts

As of 22 November 2026, per DefiLlama, total value locked is $4.7B on Base, $1.2B on Arbitrum and $369.0M on OP Mainnet, and stablecoin supply is $4.8B, $4.2B and $442.1M respectively. Base was originally built on the OP Stack and announced in February 2026 that it was moving away from it; what Base is looks at the largest of the three on its own terms.

DefiLlama’s TVL is the value deposited in protocols on the chain, and it leaves out the bridge holding the collateral. ETHBEAT publishes a different measure, total value secured, which its glossary defines as the total value that can be accessed on a project: tokens bridged through the canonical bridge, tokens bridged by other routes and tokens issued natively on the rollup. The two sites are counting different things, so the same rollup is not expected to show the same number on both.

Neither figure says anything about the applications on top. A rollup’s proofs protect the transition from one state to the next, not the contracts users deposit into. DefiLlama’s hacks record lists 106 incidents involving Arbitrum and 69 involving Base, with an incident on several chains counted on each, and how crypto gets hacked reads that record as a whole.

What to check before moving funds to one

Four questions cover most of the difference between rollups, and each has a public answer. What stage does ETHBEAT give it, and who can upgrade its contracts without notice? If the sequencer stops or ignores you, is there a forced path through Ethereum, and how long is it? Will you leave through the canonical bridge and wait out a challenge period, or through a third party that is faster and trusted in a different way? And is the data posted to Ethereum, or held somewhere else?

ETHBEAT’s project pages summarise the first, second and fourth for each rollup it tracks. The third is the user’s own choice, made at the moment of withdrawing.