Base publishes postmortem on severe transaction drops and inclusion delays

What happened

Base published a postmortem for a January 31 period of severe mainnet transaction drops, inclusion latency, and some block-production delays under high transaction load.

What changed

Base rolled back the transaction-propagation configuration change identified as the immediate cause and committed to mempool-pipeline, alerting, and change-monitoring improvements. The event added a significant realized availability failure to canonical-bridge operational diligence.

What did not change

The postmortem did not report a canonical-bridge contract exploit, asset loss, withdrawal-rule change, challenge-period change, or continuing impairment after the rollback.

Confirmed

• Base estimates that approximately 20% of submitted transactions landed on-chain during the acute incident window.
• Base reports 2.1 million dropped transactions and 512,000 transactions included in blocks.
• Approximately 2% of blocks during the reported window took longer than two seconds to build.
• Base attributed the incident to an interaction between a transaction-propagation queue-size change, rising base fees, and mempool-client behavior.
• The mitigation was to roll back the configuration change, after which Base reported restored network stability.

Still open

• The postmortem lists 19:17–20:41 UTC as the incident endpoints but calls that interval two hours and 26 minutes; those figures are arithmetically inconsistent.
• The record does not identify how many canonical-bridge deposits, proofs, or withdrawal transactions were delayed or dropped.
• Completion and effectiveness of the promised mempool-pipeline and monitoring improvements were not established by the postmortem.

What it means for an advisor

• Review the Coinbase Bridge memo's sequencer, mempool, block-builder, RPC, retry, and transaction-inclusion assumptions.
• Preserve transaction resubmission, alternate-RPC, nonce-management, and status-monitoring procedures for bridge operations during congestion.
• The resolved event does not independently fire a bridge-contract loss criterion, but recurrence should inform operational limits and exit planning.