Woofun AI reports that a consensus mechanism failure on the Base mainnet generated invalid blocks, halting new block production after block 47806542 in the early hours of June 26 Beijing time. This disruption paralyzed top-ups, withdrawals, and client software operations for approximately two hours, marking a significant operational pause distinct from previous incidents.

The event diverges from the 33-minute interruption recorded on August 5, 2025, which stemmed from a high-availability switch failure within the sorter. Despite the differing technical root causes, both outages confirm that daily L2 reliability remains critically dependent on the sorter and its associated maintenance systems. This dependency creates a single-point vulnerability that persists even as the network scales.

Compounding the operational strain, the incident occurred immediately adjacent to the scheduled Beryl upgrade window. Originally set to activate at 2:00 am on June 26 Beijing time, the Beryl mainnet launch has been postponed to 2:00 am on June 27. The delay ensures the network stabilizes before deploying critical architectural changes.

The Beryl upgrade introduces the B20 native token standard via Rust precompilation and the B20Factory singleton. This standard integrates role-based permissions, maximum supply limits, minting and burning capabilities, suspension options, transfer strategies, memo functionality, and ERC-2612 permission support to create a comprehensive token issuance toolkit.

Woofun AI data shows the upgrade also targets infrastructure efficiency by reducing the final confirmation period for one-proof withdrawals from 7 days to 5 days.

Furthermore, the implementation of Reth V2 aims to cut disk usage by up to 50% while increasing throughput by approximately 33%.

Prior to this outage, Base deployed Flashblocks to provide a pre-confirmation mechanism with a response time of around 200 milliseconds. While all mainnet blocks utilize the Flashblocks builder, applications retain the option to opt out of using these pre-confirmed datasets. The recurrence of such failures suggests that pre-confirmation mechanisms alone cannot fully mitigate systemic risks inherent in centralized sorting architectures.