Cabinet
Back to Technology Insights
2026.09.15

[Follow-up] Glamsterdam's Sepolia Fork Locked for 2026-10-06 13:53:36 UTC (Epoch 353024): First Clean Transition on Devnet-11, and the "Epoch 351232" Mismatch Explained

[Follow-up] Glamsterdam's Sepolia Fork Locked for 2026-10-06 13:53:36 UTC (Epoch 353024): First Clean Transition on Devnet-11, and the "Epoch 351232" Mismatch Explained

What happened

The public-testnet activation times for Glamsterdam were proposed by Parithosh Jayanthi of the Ethereum Foundation in ethereum/pm#2205, merged on September 15. The Sepolia values are:

  • epoch 353024
  • start slot 11296768
  • Unix time 1791294816 (Tuesday, October 6, 2026, 13:53:36 UTC)

These values were also recorded in the Meta EIP-7773 activation table via ethereum/EIPs#12355, merged September 17, and in eth-clients/sepolia:

  • EL: amsterdamTime: 1791294816
  • CL: GLOAS_FORK_EPOCH: 353024, with fork version 0x90000076

Hoodi and mainnet times remain "TBD" in pm#2205.

On September 16, Glamsterdam-Devnet-11 went through the fork transition. At ACDC #187 the next day, EF developer operations engineer pk910 reported: "The activation went perfectly fine. We saw no participation drop. All clients followed ahead." Lido and Optimism deployed their stacks on Devnet-11 and gave positive early feedback. The same call confirmed the October 6 Sepolia fork. Testing now continues mainly on Devnet-8, the public testnet Platåberget, and the Hoodi date will be set after Sepolia.

The confirmation followed a long unstable stretch:

  • Devnet-8 surfaced a consensus-layer flaw in which a block repeating its parent's hash could stop the whole network.
  • Devnet-9, launched with about 1,000 nodes, failed to finalize.
  • On the execution side, a bug in EIP-8037 (higher gas cost for state creation) forced every client to revise its implementation.

Developers skipped Devnet-10 and moved to a freshly configured Devnet-11.

The schedule has shifted several times. Sepolia activation was originally expected on August 3. A September 28 proposal emerged at ACDC #185 in August, October 6 was tentatively agreed at ACDC #186 on September 3, and it was confirmed at ACDC #187 on September 17. The ACDC #186 agenda records a request from Lido and Optimism for "at least one stable day" on a devnet before any fork is scheduled. The clean Devnet-11 transition can be read as meeting that condition. For mainnet, ethereum.org's roadmap page still says Q4 2026 with the date not yet confirmed; December has only been discussed informally among developers.

Where the number mismatch came from

This series' September 14 article noted that the "epoch 351232" reported in the press did not match "October 6, 13:53 UTC." Working from Sepolia's beacon genesis time (1655733600), epoch 351232 falls at 14:44:48 UTC on September 28, 2026, which matches the September 28 proposal from ACDC #185 on August 20. The official pm#2205 epoch, 353024, gives 1655733600 + 353024 × 32 × 12 = 1791294816, which is exactly 13:53:36 UTC on October 6. The two epochs differ by 1,792 epochs, or about 7.96 days, consistent with the Sepolia date slipping by roughly a week.

In short, coverage most likely updated the date to the new proposal while keeping the old proposal's epoch number. The epoch number is the value that goes straight into client configuration. An operator who trusted the date but set the old epoch would have a node trying to fork about eight days early.

Where compatibility breaks

Glamsterdam changes both the execution and consensus layers. The main changes are:

  • enshrined proposer-builder separation (EIP-7732);
  • block-level access lists (EIP-7928);
  • higher gas costs for state creation (EIP-8037).

The EIP-2124 fork ID changes too: Sepolia nodes will advertise 1791294816 as the next fork time in their handshakes. Erigon's implementation PR gives the post-Amsterdam fork hash as 0x6c1d9423. Older builds that do not know the fork time will drop off the network at that moment.

The impact reaches beyond full nodes. EIP-7928 adds a new block header field, so light clients that decode headers themselves, bridge header verifiers and indexer parsers also need updates. The readiness plan of myotis, an independent wallet engine, explicitly includes a pre-fork test that its forward-tolerant decoder handles Amsterdam's extra header field. Even organizations that run no nodes are affected if any of their code reads headers.

Client integration moved in parallel in late September. PRs adding the Sepolia schedule can be found for Reth, Nethermind, Erigon (r3.7 line), Lodestar, ethrex and others. However, Reth's PR #27348 was closed pending an update to a dependency (alloy-rs/hardforks#96). When each client's supporting release ships has to be checked repository by repository.

Pitfalls to watch

First, the fork ID changes early. Builds that include the fork time advertise the new forkNext in handshakes before the fork. They still connect to old builds, but the chains split the moment the fork time passes. "We're connected right now" tells you nothing.

Second, upgrading only one layer. Glamsterdam spans EL and CL, so a node that upgrades only one side will have its two layers disagree after the fork. For validators, that means missed attestations.

Third, Sepolia validates Sepolia conditions. Devnet-11's stability does not guarantee the Hoodi or mainnet schedules. Depending on how Sepolia goes, Hoodi (late October has been floated tentatively among developers) and later dates could still move. Changes that alter gas accounting, such as EIP-8037, also affect estimation logic in L2s and contracts. Sepolia is the first public environment to test that under conditions close to real traffic.

[Technical Insights & Actions]

  1. A fork date is a value for humans, but the epoch number and timestamp are values machines execute, and coverage can let them drift apart. Here the date was updated while the epoch kept the old proposal's value. [Node operators] This week, check your Sepolia settings (amsterdamTime: 1791294816, GLOAS_FORK_EPOCH: 353024) against the primary values in eth-clients/sepolia, and add to your runbook a step that re-derives the date and time from the genesis time.
  2. After nine failed devnets, a single clean transition locked in the public-testnet schedule. But stability on a devnet only proves interoperability between clients. How surrounding stacks such as L2s and staking infrastructure behave under real load is tested for the first time on Sepolia. [Integrators / L2 operators] Before the October 6 fork, prepare your L2 nodes, bridges and gas estimation logic for Sepolia, including state creation affected by EIP-8037, and staff monitoring for the 48 hours after the fork.
  3. EL and CL releases ship from separate repositories at separate times, and in some clients integration is waiting on dependency updates. [Node operators] By October 1 (five days before the fork), identify the official release versions that include the Sepolia schedule for your EL and CL pair and finish upgrading. If none has shipped, decide whether to switch to an alternative client.

[Sources]

Note: Epoch-to-time conversions were computed by the author from Sepolia's beacon genesis time of 1655733600 (2022-06-20 14:00:00 UTC), with one epoch = 32 slots × 12 seconds.

This article is a technical commentary based on publicly available information and is not investment advice.

Previous ArticleCircle Opens Arc Public Mainnet: Where a Permissioned L1 With USDC Gas Breaks EVM-Compatibility Assumptions (18 vs 6 Decimals, No WETH)Next Article [Follow-up] Ethereum and Base Abandon Effort to Unify Native Account Abstraction: EIP-8141 and EIP-8130 Proceed as Separate Standards, and Wallets Will Have to Implement Both