The Crypto Industry
Is Base Decentralized? Ethereum's Blobs, Rollup Stages, Sequencers
Ethereum raised its blob target from 3 to 14 per block in under two years, yet Base, a Stage 1 rollup, still runs a single active sequencer. What L2BEAT's stages measure, what Base's outage…

What the Rollup-Centric Roadmap Asked of Ethereum
Ethereum's base layer has delivered the data capacity its rollup roadmap asked for. The blob target, the amount of rollup data each block aims to carry, rose from 3 when blobs launched in March 2024 to 6 in May 2025, then to 10 and 14 through two small forks in December 2025 and January 2026. Every step is dated in Ethereum Foundation (EF) announcements and written into client configurations.
The rollups that buy that space have moved more slowly on another axis. On 21 October 2026, L2BEAT, whose Stages framework is the usual yardstick of rollup maturity, rated six rollups at Stage 1 and four at Stage 2, and none of its criteria asks for more than one sequencer. Base, one of the six, documents that it "currently operates a single active sequencer".
When Base's sequencer failed in August 2025 and twice in June 2026, block production stopped for 33, 116 and 20 minutes. The roadmap shipped capacity faster than it shipped the decentralization that capacity was meant to carry.
In one minute
- Blob target per block: 3 in March 2024, 6 in May 2025, 14 since January 2026, with a maximum of 21.
- Use lags capacity: about 4.8 blobs per slot in the 30 days to 20 October 2026, and 3.41 ETH in blob fees.
- L2BEAT's stages grade how safely users can exit and who controls upgrades, not whether a chain keeps producing blocks.
- Base, a Stage 1 rollup, runs one active sequencer; when it fails, the chain stops.
- Its escape hatch through Ethereum guarantees inclusion within about 12 hours, longer than any halt in Base's post-mortems.
The bargain was set out in October 2020, when Vitalik Buterin wrote on the Ethereum Magicians forum that the ecosystem was "likely to be all-in on rollups". Rollups would execute transactions away from the main chain and post the data needed to rebuild their state back to it. Base-layer scaling, he wrote, would mean "scaling how much data blocks can hold". That data is what lets anyone rebuild a rollup's state and challenge a wrong one.
EIP-4844, shipped in the Dencun upgrade, delivered the first installment. The EF's announcement set activation at epoch 269568 on 13 March 2024 at 13:55 UTC. A blob holds 4,096 field elements of 32 bytes, about 128 kilobytes, is priced in a fee market of its own and is pruned by nodes after roughly 18 days. The EIP fixed a target of 3 blobs and a maximum of 6 per block, limits it said were "expected to be increased in future upgrades as the network demonstrates reliability under larger blocks".
Pectra, 7 May 2025: Six Blobs Targeted, Nine Allowed
Pectra followed fourteen months later. The EF announced on 23 April 2025 that it would activate at epoch 364032 on 7 May 2025 at 10:05:11 UTC. Its blob change, EIP-7691, now a Final standard, raised the target and maximum to 6 and 9 blobs per block, a change the EF said "doubles Ethereum’s blob throughput".
EIP-7691 restates the premise in one line: "Ethereum, with its rollup centric roadmap, scales by relying on L2." Its authors framed the increase as short-term scaling "until future solutions are deployed", keeping client teams free for "peerDAS and other scaling research". The sampling design they meant arrived with the next upgrade.
Two other Pectra changes are often cited alongside it. EIP-7702 lets an externally owned account, the ordinary key-controlled address, delegate to smart-contract code, which wallets use for batching and gas sponsorship. EIP-7251 raised the maximum balance on which a validator earns rewards from 32 ETH to 2,048 ETH. Neither changes blob capacity.
Fusaka and Two Blob-Only Forks: From 6 to 14
Fusaka was announced by the EF on 6 November 2025 and activated at slot 13,164,544, epoch 411392, on 3 December 2025 at 21:49:11 UTC. Its headline, EIP-7594, introduced PeerDAS, "a new networking protocol that allows nodes to verify blob data availability through sampling rather than downloading entire blobs". Erasure coding lets each node check only part of the data while the network still guarantees that all of it is available.
Sampling loosened the link between blob capacity and each node's bandwidth, and Fusaka put it to work at once. Blob Parameter Only (BPO) forks, defined in EIP-7892, change nothing but the blob target, the maximum and the fee update fraction. BPO1 activated at epoch 412672 on 9 December 2025 at 14:21:11 UTC with a target of 10 and a maximum of 15. BPO2 followed at epoch 419072 on 7 January 2026 at 01:01:11 UTC with 14 and 21.
Fusaka carried other changes too. EIP-7935 set the default gas limit to 60 million, the level the EF said "core developers believe Ethereum L1 can currently safely scale to". EIP-7951 added a precompile for secp256r1 signatures, the curve used by Apple's Secure Enclave, Android Keystore and FIDO2/WebAuthn devices. EIP-7918 put a floor under the blob base fee, tied to the price of execution gas.
As of 21 October 2026, no mainnet blob increase had been scheduled since BPO2. Glamsterdam, the next named upgrade, reached the Sepolia testnet on 6 October 2026. The EF had targeted it for the first half of 2026; on 16 October 2026 it wrote that the mainnet activation time "will be announced once decided by client teams".
How Much of the New Blob Space Rollups Use
Rollups use far less space than they now have. In the 30 days from 21 September to 20 October 2026, Blobscan, a public blob explorer, counted 1,040,741 blobs, about 4.8 for each 12-second slot, against a target of 14 per block. Blobscan found blobs in 177,168 blocks over a period of 216,000 slots, so more than one slot in six passed without a blob.
Fees are correspondingly low. Blobscan puts total blob fees for those 30 days at 3.41 ETH, about 0.0000033 ETH per blob. Under Dencun the market was tight: in December 2024 blocks carried almost exactly 3 blobs on average, the target at the time, and blob senders paid 1,143 ETH in fees over the month.
Before Fusaka, the EF wrote that layer 2 usage was "often reaching the current 9 blob per block limit". Since BPO2, average use has stayed far below the new limits: from 8 January to 20 October 2026, Blobscan's daily counts average about four blobs per slot, below even the target of 6 that Pectra set. With blob space cheap and plentiful, whatever limits rollups now lies elsewhere.
What an L2BEAT Stage Measures, and What It Leaves Out
L2BEAT introduced its Stages framework in June 2023, building on milestones Buterin proposed in November 2022 for rollups "taking off training wheels". Stage 0 is a project controlled by a few entities, Stage 2 one controlled by code and Stage 1 something in between. The tests ask who can stop a withdrawal and who can change the rules.
Stage 1 rests on one principle: "The only way (other than bugs) for a rollup to indefinitely block an L2→L1 message (e.g. a withdrawal) or push an invalid L2→L1 message (e.g. an invalid withdrawal) is by compromising ≥75% of the Security Council." Upgrades by anyone outside the council need an exit window. L2BEAT's explainer still prints "at least a 7 days exit window"; the specification it calls "the source of truth" cut the minimum to five days on 30 April 2026.
Stage 2 goes further: fraud proofs open to anyone, at least 30 days to exit before an unwanted upgrade, and a Security Council that may act only on errors detected on chain. On 21 October 2026, L2BEAT's summary listed 22 rollups. Aztec Network, Cartesi PRT Honeypot v2, Ethscriptions and Facet were at Stage 2; Arbitrum One, Base, Ink, OP Mainnet, Starknet and Unichain at Stage 1; twelve at Stage 0. Another 71 chains sat under "Others", most because their proof systems are not fully functional.
Nothing in those requirements concerns how many sequencers a rollup runs. L2BEAT's project pages record that separately: Base, Arbitrum One and OP Mainnet each use "a single centralized sequencer for fast confirmations", and Starknet runs five "permissioned consensus sequencers controlled by StarkWare". All four hold Stage 1.
If the proposer set is open to anyone with enough resources, Stage 1 assumes at least one live proposer, but L2BEAT adds: "We don’t assume it to be non-censoring." Where a Stage 1 rollup answers censorship, it does so with a forced-inclusion path through Ethereum, not with a set of sequencers; a delay of hours is not an indefinite block, so a chain can halt and keep its stage. L2BEAT's own page warns that the framework "only discusses the maturity of a rollup in terms of decentralization, and not in terms of security."
The EF set the same bar for rollups in March 2026:
L2s working towards trust-minimization should at least reach Stage 1 and pass the “walkaway” test, meaning users can safely exit to L1 even in the presence of malicious operators or security council failure. — Josh Rudolf, Julian Ma and Josh Stark, authors, "How L1 and L2s can build the strongest possible Ethereum", Ethereum Foundation
The walkaway test asks whether users can leave safely. A rollup can pass it and still stop producing blocks for two hours.
When Base's Single Sequencer Stops
Base has held Stage 1 since 29 April 2025, after adding permissionless fault proofs and a Security Council. Its documentation is plain about block production: Base "currently operates a single active sequencer", which "runs as a high-availability cluster in which leadership transfers to a follower if the current leader stops producing blocks".
In the August 2025 halt, the cluster's own failover failed. At 06:07 UTC on 5 August, Base's post-mortem says, the active sequencer fell behind under on-chain activity and Conductor, the OP Stack component that manages the cluster, handed leadership to a new sequencer that "was in the process of being provisioned and was unable to produce blocks". Conductor was not yet fully enabled on that machine, so it could not hand off again. Engineers paused Conductor and moved leadership by hand; blocks stopped for 33 minutes.
The June 2026 halts began with a software fault. At 15:47 UTC on 25 June, an invalid transaction failed inside the sequencer's block builder:
An invalid transaction was received by the block builder and failed during execution, as expected, but erroneously did not clear the journal state that contained the accounts and storage slots that had been accessed. — Base Engineering Team, Postmortem: June 25th Block Production Outage, Base
The next valid transaction was charged the wrong amount of gas, and the sequencer sealed a block that validators, re-executing it, rejected. The chain stopped at block 47,806,542. Safety held, since no node accepted the bad block and Base says all funds were safe; liveness depended on one sequencer.
Recovery was slowed by a second bug: "A race condition in the engine reset feature prevented the sequencers from catching up to tip after being restarted." Base says that issue is why the incident recurred the next day, for 20 minutes. It gives the first halt as 116 minutes, although the timeline in the same post-mortem runs from 15:47 to 17:51 UTC. Earlier Base post-mortems cover halts of 29 minutes in September 2023 and 17 minutes in September 2024.
While the chain was stopped, users had the escape hatch Base's documentation describes: a transaction sent as a deposit through the OptimismPortal contract on Ethereum must be included, and if the sequencer posts nothing, nodes wait one sequencing window, 3,600 L1 blocks or about 12 hours, then derive the missing blocks themselves. The OP Stack documentation gives the same 12-hour default, and L2BEAT's page for Base calls the result "eventual censorship resistance".
Every halt in Base's post-mortems ended long before that window closed. During each one, a user could queue a deposit on Ethereum and count on its eventual inclusion, but no one could get a transaction into a Base block until the sequencer returned. In June the transaction pool overflowed and rejected new submissions.
Elsewhere in the Stage 1 table the hatch is also measured in hours: 12 for Ink, OP Mainnet and Unichain, up to a day for Arbitrum One. Starknet has none: L2BEAT says its users can submit transactions to an L1 map “but can’t force them”, and lists it as not passing the walkaway test.
Base is therefore decentralized in part. In the sense L2BEAT measures, users can exit without the operator's help, and upgrades need its Security Council. Block production is not: one operator runs the only active sequencer, and when that sequencer stops, Base stops.
What Has Shipped Toward Sequencing Without an Operator
Designs without a privileged sequencer already run on mainnet. Facet, a based rollup, takes its order from Ethereum itself: its documentation says ordering "is determined entirely by Ethereum L1, with no intermediaries", and Ethscriptions, a fork of Facet, works the same way. Aztec Network orders transactions through a sequencer randomly selected each time slot from a set that anyone can join by staking. All three are at Stage 2, beside Cartesi's PRT Honeypot v2, a bug bounty.
Base has taken another route. In September 2023, after its 29-minute halt, it wrote that the Superchain offered "the possibility of permissionless sequencing modes via modular sequencing, which would allow block building to be decentralized, removing the sequencer as a single-point-of-failures in the network." In February 2026 it announced a move off the OP Stack to "a unified, Base-operated stack"; the decentralization milestones that post names are faster withdrawals through multiple proofs, Base-specific governance and neutrality standards.
Its 2026 upgrades went to proofs, withdrawals and speed. Azul added a multi-proof system combining TEE attestations and zero-knowledge proofs, Beryl cut the single-proof withdrawal window from seven days to five, and Cobalt added validity transactions. Denim, which activated on Base Sepolia on 21 October 2026 and is planned for mainnet in November, moves the chain from one block every two seconds to five a second. None of them changes the single active sequencer its documentation describes.
On the base layer, the next dated step is Glamsterdam, whose headline changes are enshrined proposer-builder separation (EIP-7732) and block-level access lists (EIP-7928). The EF's announcement of 16 October 2026 scheduled it for the Hoodi testnet at epoch 132,352, at 17:42:48 UTC on 26 October 2026, and left the mainnet date undecided.
Sources
- Ethereum Foundation Blog: Fusaka Mainnet Announcement (6 November 2025) blog.ethereum.org
- Ethereum Foundation Blog: Pectra Mainnet Announcement (23 April 2025) blog.ethereum.org
- Ethereum Foundation Blog: Dencun Mainnet Announcement (27 February 2024) blog.ethereum.org
- EIP-4844: Shard Blob Transactions eips.ethereum.org
- EIP-7691: Blob throughput increase eips.ethereum.org
- EIP-7607: Hardfork Meta - Fusaka (activation and BPO schedule) eips.ethereum.org
- EIP-7594: PeerDAS - Peer Data Availability Sampling eips.ethereum.org
- EIP-7892: Blob Parameter Only Hardforks eips.ethereum.org
- EIP-7918: Blob base fee bounded by execution cost eips.ethereum.org
- EIP-7935: Set default gas limit to 60M eips.ethereum.org
- EIP-7951: Precompile for secp256r1 Curve Support eips.ethereum.org
- EIP-7702: Set Code for EOAs eips.ethereum.org
- EIP-7251: Increase the MAX_EFFECTIVE_BALANCE eips.ethereum.org
- Ethereum Foundation Blog: Glamsterdam Hoodi Testnet Announcement (16 October 2026) blog.ethereum.org
- Ethereum Foundation Blog: Glamsterdam Testnet Announcement (Sepolia, 6 October 2026) blog.ethereum.org
- EIP-7773: Hardfork Meta - Glamsterdam (activation table) eips.ethereum.org
- Ethereum Foundation Blog: Protocol Priorities Update for 2026 (18 February 2026) blog.ethereum.org
- Ethereum Foundation Blog: How L1 and L2s can build the strongest possible Ethereum (Josh Rudolf, Julian Ma, Josh Stark, 23 March 2026) blog.ethereum.org
- eth-clients/mainnet: mainnet consensus configuration (fork epochs and blob schedule) raw.githubusercontent.com
- go-ethereum: mainnet chain specification (blob schedule through BPO2) raw.githubusercontent.com
- Blobscan API: daily blob statistics, 21 September to 20 October 2026 api.blobscan.com
- Blobscan API: daily blob statistics since March 2024 api.blobscan.com
- L2BEAT: The Stages Framework (explainer) l2beat.com
- L2BEAT Forum: The Stages Framework (specification and changelog) forum.l2beat.com
- L2BEAT Forum: Stage 1 update, minimum challenge period reduction from 7d to 5d (30 April 2026) forum.l2beat.com
- L2BEAT: Scaling summary (stages as of 21 October 2026) l2beat.com
- L2BEAT: Base Chain project page (sequencing, Stage 1, milestones) l2beat.com
- L2BEAT: Arbitrum One project page l2beat.com
- L2BEAT: OP Mainnet project page l2beat.com
- L2BEAT: Starknet project page l2beat.com
- L2BEAT: Aztec Network project page l2beat.com
- L2BEAT: Facet project page l2beat.com
- Base documentation: Reliability and Liveness docs.base.org
- Base Status: Base Mainnet | Unsafe head delay, postmortem (5 August 2025) status.base.org
- Base Engineering Team: Postmortem: June 25th Block Production Outage (27 June 2026) base.org
- Base Status: Base Mainnet Chain Stall (25-26 June 2026) status.base.org
- Base: Base Mainnet Sep 5th Incident Postmortem (15 September 2023) base.org
- Base: Base Mainnet 09/21/24 Incident Postmortem (27 September 2024) base.org
- Base: Base has reached Stage 1 Decentralization (29 April 2025) base.org
- Base Engineering Team: A new, unified stack for Base Chain (18 February 2026) base.org
- Base documentation: Upgrades (Azul, Beryl, Cobalt, Denim) docs.base.org
- Base documentation: Beryl, Faster Withdrawals docs.base.org
- Base documentation: Denim overview docs.base.org
- Base Status: Sepolia Testnet, Base Denim Upgrade (21 October 2026) status.base.org
- OP Stack documentation: Sequencer outages docs.optimism.io
- Facet documentation: Immutable Sequencing docs.facet.org
- Aztec Network: decentralized sequencing, proving and governance aztec.network
- Vitalik Buterin: A rollup-centric ethereum roadmap (Ethereum Magicians, 2 October 2020) ethereum-magicians.org
- Vitalik Buterin: Proposed milestones for rollups taking off training wheels (Ethereum Magicians, 2 November 2022) ethereum-magicians.org