MEV, censorship, and centralized rollup sequencing are not separate problems - they all trace back to a missing protocol layer. A new ethresear.ch post by an anonymous researcher makes the case that Ethereum has been operating with three layers (consensus, execution, data) while ignoring the fourth: memory. That layer is where transactions get sequenced into rollups, mempools form, and inclusion lists either protect or betray network values.
How the trinity of layers fell short
Bitcoin’s architecture fused consensus, execution, and data availability into one replicated state machine. Ethereum’s proof-of-stake explicitly separated consensus from execution - the consensus block now wraps the execution block. Then rollups and data blobs forced a third data layer into the protocol. But the post argues that rollups’ off-chain sequencers created a new territory: the memory layer. Sequencers decide which transactions go into rollup blocks. They control the mempool. They can censor. And the L1 protocol has zero purview over that decision.
The problem is that Ethereum’s scaling roadmap treated execution and data as the only frontiers. The memory layer - the ordering and inclusion of transactions - was left to centralized sequencers and private mempools. The result: MEV extracted by builders, censorship by sequencers, and a growing gap between L1 ideals and rollup reality.
FOCIL, sharded mempools, and the memory layer toolset
Once you frame these issues as memory-layer problems, the existing proposals snap into focus. The post lists them explicitly: FOCIL (inclusion lists), sharded mempools (also called based sequencing), gas futures, frame transactions (account abstraction), encrypted mempools, and private mempools. These aren't disparate features - they are mechanisms for governing transaction sequencing under protocol purview.
The endgame is a virtual, sharded mempool that turns mainnet into “one big sequencer.” Transactions would be sequenced into native rollup-like execution shards with dynamic throughput allocation. Synchronous composability across sub-shards becomes real. Censorship becomes much harder because no single sequencer holds veto power. Toxic MEV gets starved by protocol-enforced ordering rules.
What changes if we acknowledge the fourth layer
Admitting the memory layer exists reshapes the research agenda. Instead of fighting MEV with ad-hoc patches, we build a unified theory of transaction sequencing. Instead of trusting rollup sequencers, we enforce inclusion through FOCIL or sharded mempools. The post even suggests a strawmap of research topics: FOCIL, sharded mempools, base sequencing, gas futures, frame transactions, encrypted mempools, private mempools - all parts of the same layer.
Ethereum’s path to credible neutrality requires bringing this layer under protocol homogenization - just as execution sharding and data availability sampling did for their domains. The goal: fair inclusion, predictable ordering, and minimally extractive sequencing. The memory layer turns the infinite garden of MEV and censorship into a set of well-defined protocol mechanisms we can actually reason about. The next step is building the virtual mempool that makes mainnet the sequencer for all rollups.
Source: Ethereum's Fourth Protocol Layer: Memory
Domain: ethresear.ch
Comments load interactively on the live page.