ACDT #101 · Builder deposit fix excluded Hoodi fork stays on schedule
AllCoreDevs - Testing #101 · 2026-10-19
They decided to keep the testnet and mainnet schedule rather than make a last-minute consensus change
They decided not to include the fix separating builder and validator deposits in Glamsterdam, and to fork the Hoodi testnet on October 26 as scheduled. The EL's BAL retention period was also fixed at 14,299 epochs, so CL clients can proceed with blinding.
Why it matters
The Glamsterdam mainnet upgrade is now more likely to stay on schedule for this year. However, staking services such as Lido will have to manage the deposit attack risk with their own workarounds for the time being.
Decisions
- Builder deposit fix dropped: The fix that separates builder deposits from validator deposits will not be included in Glamsterdam. The judgment is that this is a hypothetical attack and not worth delaying the Hoodi fork.
- BAL retention 14,299: The PR (EIPs#12347) raising BAL retention to 14,299 epochs will be merged, and CL clients will proceed with blinding. The discussion on reducing it to 8,192 epochs was postponed to Hegotá.
- devnet-2 quick slots dropped: heze-devnet-2 will not include quick slots and will contain only the EIPs merged in Beta-4. This means the short slot experiment is put on hold for now.
Discussion
Should the builder and validator deposit separation fix go in before Hoodi?
It was pointed out that under the current builder onboarding logic an attacker can redirect a validator deposit to a builder they control, and whether to change the spec one week before the fork became the point of contention.
Outcome: Decision not to include in Hoodi 00:22:53
How long should BAL be retained
This discussion decides how long the EL must keep BALs so that it can still answer requests for historical payloads after the CL blinds the BALs.
Outcome: PR merge decision 00:37:23
Can an EIP be merged into the main spec branch before SFI?
A governance debate arose over putting quick slot code, which has not yet been Scheduled for Inclusion (SFI), into the Heze fork branch of the consensus spec, with the concern that sunk costs lead to implicit scheduling for inclusion.
Should heze-devnet-2 include 10-second slots?
As CL clients implementing short slots reached the agreed threshold of three, the discussion covered whether to shorten the slot time in the next devnet and how to align EL blob parameters.
Outcome: Decision to exclude quick slots 00:55:56
Glossary
- SFI: Scheduled for Inclusion, the stage where inclusion in the fork is confirmed
- CFI: Considered for Inclusion, the stage where inclusion in the fork is under review
- PFI: Proposed for Inclusion, the stage where inclusion in the fork has been proposed
- DFI: Declined for Inclusion, the state of being removed from inclusion candidacy
- BAL: Block Access List, a list of the state data a block touched, which clients retain for a set period
- Blinding: blinding, removing the data body and keeping only a reference to reduce storage space
- BPO: Blob Parameter Only fork, a small fork that changes only the blob capacity parameters
- Quick slots: quick slots, a proposal to shorten the slot time, the block production interval, to less than 12 seconds