Glossary · Scaling

Rollups

Published 23 May 2025 · Updated 2 Nov 2026 · 13 min read

A rollup is a blockchain that executes transactions outside its base chain, usually Ethereum, and posts data and new state roots back to it. L1 contracts accept results through fraud proofs for optimistic rollups or validity proofs for ZK rollups. Anyone can rebuild its state from the posted data.

  • Since March 2024, rollups have mostly posted compressed transaction data or state differences in blobs.
  • In November 2026, withdrawals take about 5 to 7 days on optimistic rollups and hours to a day on ZK rollups.
  • On 2 November 2026, L2BEAT lists 12 of its 22 rollups at Stage 0, with training wheels still in place.
How a rollup worksOptimistic and ZK rollups comparedWhere rollups publish their dataRollup stages and training wheelsForced inclusion and exit windowsRollup risks to understandHow rollups developedDecisions before building a rollupRollups FAQ

How a rollup works

Transactions run outside Ethereum, where processing them can cost less. Data and results return to Ethereum for checking or challenge. In this layer 2 design, a sequencer orders transactions into a batch. A proposer submits a state root, which commits to the resulting state. The rollup contract accepts that result under its proof rules.

Number 1

Sequence transactions

The sequencer orders and executes each L2 transaction. Every transaction runs off chain, including those on an optimistic rollup that nobody challenges.

Number 2

Post the batch

The batch poster compresses the data and sends it to Ethereum. Publication uses calldata or blobs, rather than a summary of results.

Number 3

Propose a root

The proposer submits a state root to the L1 contract. This commitment represents the new blockchain state, not the data needed to reconstruct it.

Number 4

Prove or challenge

An optimistic rollup accepts a claim unless challenged through a fraud proof. A ZK rollup submits a validity proof for a batch or several batches.

Number 5

Settle and withdraw

The contract accepts the result after the proof rules are satisfied. A withdrawal to Ethereum must also satisfy the system's challenge period or execution delay.

Rollups publish compressed transaction data or state differences, enough for anyone to rebuild the state. In January 2021, Vitalik Buterin illustrated compression with ETH transfers: about 110 bytes on Ethereum versus 12 in a rollup. IPFS cannot replace L1 publication because it provides no consensus on data availability. Challengers cannot construct fraud proofs without execution inputs, and users cannot know balances when state details are withheld. Plasma's data model moves data off chain, creating a mass exit problem when an operator withholds it.

The OP Stack's output root combines the state root, withdrawal storage root and latest block hash. On 2 November 2026, Arbitrum One validators post an assertion about every hour. Cannon is the OP Stack's default fault proof virtual machine on that date. L2BEAT requires a source-available node that reconstructs state from L1 data. A commitment does not make transaction details confidential. Rollups are not private by default, even when their proof systems use zero-knowledge techniques.

Optimistic and ZK rollups compared

The distinction is how Ethereum accepts a result, not whether transactions execute off chain. An optimistic rollup allows challenges to claimed results. Its safety requires one honest node. A ZK rollup supplies a proof of correct execution. Both L2 designs publish reconstruction data. Timings reflect documentation read on 2 November 2026.

Optimistic rollupsZK rollups
How a result is acceptedOptimistic rollupsAccepted after a challenge period unless successfully disputedZK rollupsAccepted after verification of a validity proof
What Ethereum checksOptimistic rollupsOne disputed execution step after bisectionZK rollupsA cryptographic proof, such as a ZK-SNARK proof
Withdrawal to EthereumOptimistic rollupsAbout 5 days on Base to 7 days on OP MainnetZK rollupsHours to a day, depending on proof and safety delays
Main cost besides dataOptimistic rollupsAbout 40,000 gas per batch in Vitalik Buterin's 2021 estimate, plus proposer bondsZK rollupsAbout 500,000 gas to verify each proof, plus proof generation
ExamplesOptimistic rollupsArbitrum One and OP MainnetZK rollupsZKsync Era, Starknet, Scroll and Linea
Figure 1 · Documentation and L2BEAT, read on 2 November 2026Follow an exit to EthereumCompare five networks from one second to two weeks. Toggle a dispute or censorship to see added waiting.
Focus
What if
Time until funds are usable on Ethereum, log scaleOptimistic rollupZK rollupFast exitForced inclusion waitStep
Optimistic rollups6.4 days to 7 daysArbitrum One and OP Mainnet
ZK rollups2 h to 1 dayZKsync Era, Linea and Taiko
Fast exitSecondsPaid to a liquidity provider, who waits instead
Optimistic rollups7 days to 14.7 daysArbitrum One and OP Mainnet
ZK rollups2 h to 1 dayZKsync Era, Linea and Taiko
Fast exitSecondsPaid to a liquidity provider, who waits instead
Optimistic rollups7.4 days to 7.5 daysArbitrum One and OP Mainnet
ZK rollups4.2 h, or no deadlineZKsync Era, Linea and Taiko
Fast exitAfter inclusionPaid to a liquidity provider, who waits instead
Optimistic rollups7.5 days to 15.7 daysArbitrum One and OP Mainnet
ZK rollups4.2 h, or no deadlineZKsync Era, Linea and Taiko
Fast exitAfter inclusionPaid to a liquidity provider, who waits instead

Arbitrum One and OP Mainnet withdrawals wait about 6.4 and 7 days, respectively.

Live ZK rollups settle in hours to a day because proofs and safety delays take time.

A liquidity provider can pay out on Ethereum within seconds, taking over the wait for a fee.

BoLD bounds a disputed Arbitrum One withdrawal at about 14.7 days, including the security council grace period.

Forced inclusion adds up to 24 hours on Arbitrum and 12 on OP Mainnet. Some ZK rollups lack this route.

Times follow Arbitrum, Optimism, ZKsync, Linea and Taiko documentation and L2BEAT, read on 2 November 2026.

In February 2025, L2BEAT described a week as time for a social response to a 51% consensus attack, not exchanging dispute moves. On 2 November 2026, BoLD bounds Arbitrum disputes at two challenge periods plus a two-day security council grace period. ZK rollups still need proof time and safety delays. L2BEAT reports ZKsync Era's execution delay as one day since 8 October 2026, while its documentation still says three hours.

Where rollups publish their data

L1 data lets independent participants check state, a core concern of modular blockchain design. Dencun introduced EIP-4844 on 13 March 2024, and several rollups switched within two days. Its 128 KiB blobs have a separate fee market and are pruned after about 18 days. Fusaka added PeerDAS sampling on 3 December 2025. Since BPO2 on 7 January 2026, Ethereum targets 14 blobs per block, with a maximum of 21.

Figure 2 · Page team calculation, prices of 2 November 2026Compare the cost of dataChange transaction bytes and the blob price. Compare calldata charges, blob data and the fixed cost of posting a batch.
Blob price
Focus
100 bytes
L1 data cost of one transactionUS dollars per transaction, log scale
A full blob as floor calldata737×Times its blob fee, with the 21,000 gas of the transaction
Blob data plus batch share$0.0000113.7% of a $0.0003 Base transfer
One full blob$0.0037Blob fee only, 2 Nov, 18:44 UTC

100 calldata bytes cost about $0.0021 at the floor. A full blob's calldata equivalent costs 737 times its blob fee.

The 29 October 2026 daily average blob gas price was about 26 times the 18:44 UTC sample on 2 November 2026.

At 18:44 UTC, 100 bytes in a blob cost about $0.000003, excluding execution and batch overhead.

At these prices, 100 blob bytes cost about 1% of Base's $0.0003 median transfer fee on 30 October 2026.

A batch transaction's fixed 21,000 gas costs about $0.011, exceeding the $0.0037 fee for one blob.

The page team calculates costs using Ethereum prices read on 2 November 2026 and EIP-4844, EIP-2028 and EIP-7623 gas rules.

The data charge is only part of a transaction fee. On 2 November 2026, OP Stack fees combine an execution gas fee, an L1 data fee and an operator fee. Arbitrum separates payments to its batch poster from the network fee for L2 resources. ZK rollups also pay for proof generation and verification. Calldata costs 16 gas per non-zero byte under EIP-2028. Since Pectra, EIP-7623 applies a 40-gas floor per non-zero byte to data-heavy transactions.

Rollup stages and training wheels

Vitalik Buterin proposed training wheels milestones in November 2022. L2BEAT introduced its stages in June 2023 to describe decentralization maturity, not security. A higher stage means fewer operator controls under the framework, not a guarantee against bugs or losses.

Figure 3 · L2BEAT stage code, 2 November 2026Check the remaining training wheelsTick requirements or choose a preset. See the highest completed stage and the missing requirements in L2BEAT's own words.
Preset
Stage 0 requirements
Stage 1 principle and requirements
Stage 2 requirements
ResultNot a staged rollupTraining wheels onLimited training wheelsNo training wheelsNo stageStage 0Stage 1Stage 2Missing for Stage 0, in L2BEAT's words:Missing for Stage 1:Missing for Stage 2:Every requirement in L2BEAT's stage code is met.Upgrades unrelated to onchain provable bugs provide at least 30d to exit.The Security Council is limited to acting solely on onchain provable bugs.
Stage 2 on L2BEAT4 rollupsSecuring about $0.6 million
Stage 1 on L2BEAT6 rollupsSecuring $26.62 billion
Stage 0 on L2BEAT12 rollupsSecuring $3.15 billion

Stage 1 is met, but upgrades provide less than 30 days to exit, or another Stage 2 requirement remains unmet.

A starting requirement is missing, so no stage applies. Complete every Stage 0 requirement before testing later groups.

The principle fails if, excluding bugs, actors without 75% council approval can indefinitely block withdrawals or force invalid ones with a sub-seven-day exit window.

Every requirement is met. The rollup reaches Stage 2, where council intervention is restricted to onchain provable bugs. Optimistic rollups also allow permissionless fraud proofs.

The rollup meets every Stage 0 requirement, but a Stage 1 requirement remains unmet. Training wheels still constrain independent operation.

Requirements and wording follow L2BEAT's stage code as applied on 2 November 2026.

On 2 November 2026, four rollups are at Stage 2, six at Stage 1 and twelve at Stage 0. Stage 1 rollups secure about 89% of rollup value. Since 17 August 2026, proving requirements cover published prover code, no red-rated trusted setups, and verifiers and programs regenerable from source. A qualifying security council needs at least eight public members, a 75% threshold and at least half outside the core organization. Users must also be able to exit without the operators.

Forced inclusion and exit windows

An escape hatch lets any user exit with assets under any circumstance. Forced inclusion gets a transaction past a censoring sequencer through L1. An exit window is time to leave before an unwanted upgrade. The table follows L2BEAT on 2 November 2026.

RollupHow a censored user gets a transaction inExit window or upgrade delay on L2BEAT
Arbitrum OneHow a censored user gets a transaction inSubmit through the delayed inbox, then force inclusion after at most 24 hoursExit window or upgrade delay on L2BEAT10-day regular exit window
OP MainnetHow a censored user gets a transaction inSend through OptimismPortal for inclusion within the 12-hour sequencing windowExit window or upgrade delay on L2BEATNo regular exit window
ZKsync EraHow a censored user gets a transaction inQueue transactions on L1, but inclusion cannot be forcedExit window or upgrade delay on L2BEAT5-day regular exit window
StarknetHow a censored user gets a transaction inNo user-callable forced-inclusion function is availableExit window or upgrade delay on L2BEAT7-day regular exit window
ScrollHow a censored user gets a transaction inEnforced mode activates after seven days without finalizationExit window or upgrade delay on L2BEATNo regular exit window
LineaHow a censored user gets a transaction inUsers cannot force any transaction through L1Exit window or upgrade delay on L2BEATNo regular exit window
TaikoHow a censored user gets a transaction inQueued transactions become due after 9 minutes 36 seconds, but cannot be forcedExit window or upgrade delay on L2BEATNo regular exit window

On 2 November 2026, Base streams 200-millisecond Flashblocks as soft confirmations. Taiko describes itself as a based rollup, with preconfirmations in about 0.5 to 2 seconds. Astria's shared sequencer network was halted on 1 December 2025. Arbitrum batches become final on Ethereum in about 12 to 15 minutes. A short block time measures block production, not settlement or asset release. Soft confirmations are promises, not settlement.

Rollup risks to understand

Proofs and published data do not remove every trust assumption. Operators can stop, contracts can contain bugs, and administrators may retain powerful keys. A bridge contract also needs scrutiny alongside execution and data availability. The examples below predate 2026.

Sequencer outages

Base stopped producing blocks for 29 minutes on 5 September 2023. A sequencer interruption can delay transactions even when Ethereum continues operating.

Proof system bugs

In September 2023, ChainLight found a soundness bug in ZKsync Era's circuits before any loss. Invalid execution can pass an incorrectly implemented proof system.

Upgrade key control

A malicious contract upgrade can steal funds. Without an exit window, users may have no time to leave before new code takes effect.

Transaction censorship

In June 2024, Linea paused its sequencer and censored attacker addresses after a decentralized exchange on the network was exploited. It later said its team would lose that power.

Bridge contract bugs

In September 2022, a researcher found an Arbitrum delayed inbox bug that could hijack ETH deposits. The report earned a 400 ETH bounty before any loss.

Data retention limits

Nodes need published data while it is available to rebuild state. Blob pruning after about 18 days makes availability different from permanent L1 storage.

How rollups developed

The design develops from earlier attempts to move execution away from Ethereum without losing a verifiable connection to it. The milestones below track data availability, public launches, proof systems and the removal of training wheels, rather than token performance.

  1. August 2017

    Plasma explores off-chain data

    Joseph Poon and Vitalik Buterin published the Plasma paper. Its off-chain data approach preceded rollups, which keep reconstruction data on L1.

  2. August 2019

    Optimistic rollups get their name

    John Adler described Minimal Viable Merged Consensus in June 2019. In August 2019, Karl Floersch of Plasma Group called the design an optimistic rollup. Execution moves off chain while data stays on L1.

  3. December 2019

    Loopring reaches Ethereum mainnet

    Loopring 3.0 deployed on Ethereum mainnet in December 2019. The next year brought another launch, with zkSync 1.0 going live in June 2020.

  4. October 2020

    A rollup roadmap for Ethereum

    Vitalik Buterin published a rollup-centric roadmap for Ethereum. It placed off-chain execution and onchain data within the base chain's scaling direction.

  5. August 2021

    Arbitrum One opens publicly

    Arbitrum One opened to the public on 31 August 2021. Starknet Alpha followed with its mainnet launch in November 2021.

  6. March 2024

    Dencun adds a data market

    Dencun introduced blobs on 13 March 2024. Starknet, ZKsync Era, Arbitrum, OP Mainnet and Base switched to them within two days.

  7. February 2025

    BoLD activates on Arbitrum

    BoLD went live on Arbitrum One and Nova on 12 February 2025. Its dispute process bounds delays rather than removing the challenge period.

  8. August 2026

    Proving requirements become stricter

    Four new Stage 1 proving requirements took effect on 17 August 2026. They address published code, trusted setups and regenerating verifiers and programs from source.

Decisions before building a rollup

Building a network and launching an application on an L2 involve different choices. We help frame those choices around execution, proof timing, data publication and control. Our ZK rollup development work covers custom zkEVM stacks. Our blockchain development services also cover custom Layer 1 and Layer 2 builds.

PixelPlex case icon

PixelPlex was founded in 2007, and we have built blockchain systems since 2013. Our Echo protocol case study describes a ZK-rollup bridge that executes transactions on Echo and posts a validity proof to Layer 1. Its contract keeps transfer data. Echo also has its own consensus and committee-run sidechains to Ethereum and Bitcoin.

  1. Stack and execution

    We consider the virtual machine and stack together. Our ZK rollup development experience spans ZKsync, Polygon zkEVM, Starknet and Scroll.

  2. Proofs and delays

    We assess ZK-SNARK and STARK choices, including trusted setups. A recursive proof can finalize several blocks, while Ethereum verification costs about 500,000 gas. Proving time adds delay.

  3. Data publication

    We compare blobs with calldata and account for compression, batch overhead and proof costs. The data charge alone does not describe the user's total fee.

  4. Sequencing and inclusion

    We separate sequencer confirmations from settlement and examine forced inclusion. A queue on L1 is not necessarily a way to force execution.

  5. Upgrades and governance

    We examine upgrade authority and the intended stage. Stage 1 requires at least seven days to exit before upgrades by less decentralized actors.

Discuss the requirements behind your rollup design

We can review execution, proofs, data and governance with your team before implementation choices are fixed.

Discuss your requirements

Rollups FAQ

What is a rollup in simple terms?

A rollup is a blockchain that executes transactions outside a base chain, usually Ethereum, while publishing data and results there for independent verification. Its proof system enables a trust-minimized bridge. Without a bridge, or when the bridge is not considered, L2BEAT calls the design a sovereign rollup.

How do optimistic and ZK rollups differ?

An optimistic rollup leaves a result open to challenge, while a ZK rollup proves correct execution. Both run transactions off chain. On 2 November 2026, L2BEAT classifies a system as optimistic if it can accept state roots optimistically, even when it includes a ZK component.

Why do optimistic rollup withdrawals take seven days?

An optimistic rollup's challenge period allows a response to disputed results and potential consensus attacks, rather than simply timing a fraud proof exchange. On 2 November 2026, OP Mainnet's seven days comprise a 3.5-day game clock and a 3.5-day finality delay. A faster bridge payout transfers the wait to another party. Base shortened its usual window to five days on 25 June 2026.

Are all layer 2 networks actually rollups?

No. A validium network or an optimium stores data outside a permissionless L1, so it is not a rollup. Sidechains also do not derive their security or data availability from the main chain. The layer 2 label alone does not identify a system's proof or data model.

Do ZK rollups keep all transactions private?

No. A ZK rollup's scaling proofs establish correct batch execution but do not make its published transaction data confidential. On 2 November 2026, ZKsync Era and Starknet publish state differences, while Linea publishes transaction data.

What happens if a rollup sequencer goes down?

Transaction ordering may stop until the operator recovers or a supported alternative path works. An Ethereum inbox only helps if the system lets users force inclusion. On 2 November 2026, ZKsync Era permits L1 queuing but does not let users force those queued transactions.

What do rollup stages on L2BEAT measure?

L2BEAT's stages measure decentralization maturity, not security. On 2 November 2026, L2BEAT shows stages for rollups and for validiums and optimiums, but not for projects categorized as Others. A project's classification therefore needs to be read alongside its proof system, data availability and upgrade controls.

Related terms

All definitions
Sources (35)

Get updates about blockchain, technologies and our company

We will process the personal data you provide in accordance with our Privacy policy. You can unsubscribe or change your preferences at any time by clicking the link in any email.

Follow us on social networks and don't miss the latest tech news

  • facebook
  • X
  • linkedin
  • instagram
Stay tuned and add value to your feed