Summary
The next generation Ethereum consensus decouples the dynamically-available chain (“heartbeat”) from finality, addressing both Gasper’s theoretical weaknesses and practical demands for faster confirmation, PQ readiness, and sub-256-validator slots. Goldfish provides the dynamically-available layer (safe and live as long as most awake validators are honest); Majorum adds a stabilizing finality gadget. The two-layer architecture is a fundamental impossibility result: no single protocol can be both dynamically available and provide finality.
Why Change Gasper?
Gasper = LMD-GHOST + Casper FFG. Problems:
- LMD-GHOST is not provably dynamically available: adversarial strategies can violate both safety and liveness in the synchronous model (Neu, Tas, Tse 2021).
- Complex interaction: LMD-GHOST and FFG interact in ways that are hard to reason about formally; multiple patches have not fully closed known attack vectors.
- Slot time bottleneck: current 12-second slots require aggregating ~30,000 attestations through multiple aggregation rounds. Reducing to the full validator set (1M+) under SSF would make this worse.
- PQ challenge: BLS signature aggregation has no PQ equivalent at practical sizes; ~256 validators per slot makes naive concatenation feasible.
The Availability-Finality Dilemma
A fundamental impossibility result (blockchain CAP theorem):
No single protocol can simultaneously guarantee:
- Liveness under dynamic participation: chain continues to grow as the awake validator set fluctuates
- Safety under network partitions: confirmed transactions cannot be reverted even if the network temporarily splits
BFT protocols (PBFT, Tendermint, HotStuff) achieve partition safety but halt when >⅓ validators go offline. Longest-chain protocols (Bitcoin) achieve dynamic availability but offer only probabilistic confirmation. The two-layer architecture is mathematically necessary.
Why Dynamic Availability Matters
Ethereum has never stopped producing blocks, even through:
- May 2023: Prysm/Teku bugs caused first mainnet inactivity leak. >90% of slots still proposed.
- Nov 2020: Geth consensus bug; small fork, but chain continued.
- Real-world outages push participation as low as ~33% — below any BFT fault threshold.
Dynamic availability ensures:
- Resilience: bugs/outages don’t halt the chain; recovery is “rejoin, don’t restart”
- Censorship resistance: a minority can continue building an alternative fork under majority censorship
- Application continuity: DeFi, rollups, and bridges depend on a functioning L1 — a halt simultaneously freezes all composable protocols
Compare: Solana’s Feb 2024 halt (5 hours, DeFi completely frozen) vs. Ethereum’s May 2023 (no block production halt).
Goldfish Protocol
Goldfish is a propose-and-vote protocol designed for dynamic availability:
Key Properties
- Constant expected confirmation latency: confirmation time does not grow with target security level (key separation from longest-chain protocols)
- Reorg resilience: blocks by honest proposers remain canonical — eliminates a class of attacks plaguing LMD-GHOST
- Subsampling: ~256 validators per slot, small enough for a single subnet broadcast; no aggregation rounds needed
- Composability: serves as the dynamically-available component in an ebb-and-flow protocol
Mechanism (Simplified)
- Each slot: a random proposer broadcasts a block.
- A subsampled committee (~256 validators) votes for the block.
- Vote expiry: old votes don’t accumulate indefinitely (unlike LMD-GHOST’s “ghost” votes) — prevents long-range attacks.
- View-merge: validators synchronize their local views upon receiving new blocks, ensuring consistent fork-choice.
- A block is confirmed when it accumulates sufficient votes in the current view.
3-Step Fork-Choice
- Validate block and check vote count
- Apply view-merge rule
- Select the head with the most recent sufficient quorum
Majorum: The Finality Gadget
Majorum stabilizes the dynamically-available heartbeat by providing irreversible finality:
Design
- Runs in parallel with Goldfish, not on the critical path for block production
- Uses the full active validator set for finality votes (currently ~1M validators)
- Finalizes checkpoints in the Goldfish chain after sufficient votes accumulate
- Inactivity leak: if finality stalls, inactivity penalties gradually reduce the stake of non-participating validators until finality can resume with the remaining stake
- During finality stalls, the Goldfish heartbeat continues uninterrupted
Properties
- Strictly accountable safety: ≥⅓ of stake must be slashed for a finality reversion
- Liveness: if ≥⅔ awake stake is honest, finality resumes
- Decoupled from heartbeat: a finality stall has no effect on block production
Target Architecture
┌─────────────────────────────────────────────────────┐
│ Heartbeat: Goldfish (~256 validators/slot) │
│ - Dynamic availability │
│ - ~200ms–1s sub-slot confirmations (with TOOL) │
│ - PQ-ready: 256 concatenated signatures │
└──────────────────────┬──────────────────────────────┘
│ feeds
┌──────────────────────▼──────────────────────────────┐
│ Finality: Majorum (full validator set) │
│ - Trails heartbeat by N checkpoints │
│ - Off critical path; inactivity leak for recovery │
│ - BLS signatures today; PQ aggregation in future │
└─────────────────────────────────────────────────────┘
Post-Quantum Benefits
With ~256 validators per slot:
- Signatures can be naively concatenated (no aggregation needed)
- 256 signatures at ~3 KB each = ~768 KB total
- leanMultisig (zkVM targeting XMSS aggregation) can compress to ~300–500 KB
- This means a post-quantum heartbeat can be deployed before PQ aggregation for the full validator set is solved
- The two-layer architecture converts PQ migration from an all-or-nothing problem into two independent sub-problems
Leaderless BFT (Longer Term)
Beyond Goldfish, leaderless BFT completely removes the proposer/leader role:
- No designated block proposer — any validator can propose transactions
- Blocks emerge from Byzantine consensus without a leader
- Eliminates timing games (no proposer has a privileged position to delay) and removes the primary motivation for PBS
- Building blocks: Reliable Broadcast, Binary Byzantine Agreement (BBA), Common Coin
- Representative protocols: HoneyBadgerBFT, Dumbo, Shoal++
MEV implication: without a designated block proposer, the current PBS model (proposer sells block production rights to a builder) no longer makes sense. MEV auctions would need to be redesigned for committee-level competition.
Open Questions
- What exact slot time does the Goldfish/Majorum architecture target? (Vitalik proposed: 12 → 8 → 6 → 4 → 3 → 2 seconds, gradually by √2 factor)
- Can the inactivity leak mechanism be simplified for the decoupled architecture?
- How does the 256-validator committee selection interact with PeerDAS data custody responsibilities?
- When does Native DVT (protocol-embedded DVT) make it into the roadmap?
Related Pages
- Finality in Ethereum: Gasper, Gloas, and the Engine API — Current finality in Ethereum; Gloas nuances
- Fast Confirmation Rule (FCR) — FCR as an interim solution before Goldfish
- Ethereum Protocol Roadmap 2026 — Strawmap; slot time reduction path
- Timing Games and Proof-of-Time — How PBS timing games disappear with leaderless BFT
Key Sources
- Unblocking Faster Finality with Decoupled Consensus (2026) — Goldfish code + formal proofs; Majorum; 3-step fork-choice
- Why Ethereum Needs a Dynamically Available Protocol (Luca Zanolini, Mar 2026) — impossibility result; Solana halt comparison; PQ benefit; Goldfish properties
- Preface Research for Leaderless BFT Protocol Designs (2026) — leaderless BFT survey; MEV/timing game motivation