Research
Wonderboom -- Efficient, and Censorship-Resilient Signature Aggregation for Million Scale Consensus
Ethereum is secured by almost one million validators, and this very scale makes finality slow — roughly 15 minutes — because aggregating and disseminating ≥ 2N/3 signatures across that many…

- arXiv
- 2602.06655
- Published
- 2026-02-06
- Authors
- Zeta Avarikioti, Ray Neiheiser, Krzysztof Pietrzak, Michelle X. Yeo
AI summary
Overview
Research area: Distributed systems and blockchain consensus security (cs.CR) — specifically signature aggregation and censorship resistance for large-scale Proof-of-Stake validator sets.
Technical level: Advanced. The paper assumes familiarity with Byzantine fault tolerant consensus, BLS signature aggregation, gossip networks, and Ethereum's epoch/slot/committee structure.
Scope: The paper designs, analyzes, and simulates Wonderboom, a tree-based signature aggregation protocol that aggregates all N validator signatures within a single Ethereum slot while providing stronger, formally defined vote-censorship resilience than Ethereum's current design.
What This Paper Is About
Ethereum is secured by almost one million validators, and this very scale makes finality slow — roughly 15 minutes — because aggregating and disseminating ≥ 2N/3 signatures across that many participants is expensive in computation and bandwidth. The authors show that Ethereum's existing aggregation protocol also has exploitable weaknesses that let adversaries shift stake from honest to adversarial nodes, and they propose Wonderboom: a deeper, tree-based aggregation protocol that aggregates every validator's signature every slot, enabling two-slot finality, with explicit inclusion (vote-censorship) guarantees.
Key Contributions
-
Wonderboom protocol: A censorship-resilient tree-based aggregation protocol that aggregates signatures from millions of validators within one slot, designed to be modular so it can be integrated into existing large-scale systems such as Ethereum with minimal changes.
-
Theoretical analysis of Ethereum's vote-censorship resilience: The authors analyze Ethereum's current aggregation protocol, argue that the widely used committee-composition metric is the wrong one, and show that a representation-based metric yields censorship probabilities several orders of magnitude higher — meaning existing guarantees are significantly overstated.
-
First million-scale simulation framework: A simulator capable of evaluating worst-case performance of large-scale Byzantine fault tolerant signature aggregation, including the worst case of N/3 faulty nodes.
-
Evaluation: Using the simulator, Wonderboom aggregates N signatures over 4x faster than the Ethereum design while providing superior censorship resilience; combined with proposer–validator separation this yields a total speed-up of over 32x.
Main Findings
-
Finality latency: Ethereum takes around 15 minutes to finalize a block; Wonderboom aggregates all N signatures each slot, enabling two-slot finality of approximately 24 seconds.
-
Raw aggregation speed: Wonderboom aggregates N signatures over 4x faster than the Ethereum design, and in the worst case can aggregate and verify more than 2 million signatures within a single Ethereum slot.
-
Committee-safety metric is the wrong metric: For a committee of n = 128 drawn from N = 1,000,000, the probability of a 2/3 Byzantine supermajority is about 5.55 × 10⁻¹⁵ — but this concerns committee composition, not representation. The probability that all selected representatives are faulty is approximately 1.45 × 10⁻⁵, several orders of magnitude larger.
-
Censorship events are a daily occurrence: Aggregated over 7,200 slots per day and at least 64 committees per slot, the daily probability that at least one committee experiences representative-driven censorship is approximately 0.9987, corresponding to an expected value of over 6 committees per day.
-
Proposer censorship is the dominant risk: With f ≈ N/3 Byzantine power, the proposer is Byzantine with probability ≈ 1/3 in any given slot; the paper notes two entities already own over 30% of total stake.
-
Economic damage estimate: Under a reward loss factor of 2/27 and a representative per-attestation reward of $0.03 (about $2,465 per year under full participation), honest validators lose roughly $205 annualized each; with a conservative baseline of 1 million validators and 2/3 honest stake, aggregate annual damage exceeds $136M — arising from inclusion failures, not safety violations.
-
Proposer/validator separation: Splitting the proposer process and validator process onto distinct IP addresses reduces the attack surface for deanonymization via the gossip layer, which prior work has shown is straightforward; gossip is retained only for the two phases that involve the proposer (block broadcast and delivery of final aggregates).
-
Forwarding rule against vote censorship: Combining the largest aggregate with one uniformly random aggregate, paired with a once-per-k reward allocation, lower-bounds the per-slot inclusion probability p; the chance of missing rewards is (1−p)^k, which declines exponentially in k.
-
Better resilience at smaller k: For k = 64 (two epochs) in Wonderboom, the authors' analysis indicates Ethereum would need to extend its effective penalization window to six epochs to reach similar resilience — i.e., Wonderboom allows a k three times smaller than Ethereum.
-
Latency budget: Wonderboom has approximately 4 seconds to verify the previous block, verify and aggregate N signatures, and transmit aggregates to the next proposer, given that gossip alone takes up to ≈4 seconds to reach ≈98% of validators and there are two gossip phases in a 12-second slot.
Methodology in Plain English
The authors first break down how Ethereum's aggregation actually works: validators are split into committees of N/32, each divided into 64 subcommittees of at least 128 and at most 2048 validators, with 16 randomly chosen representatives per committee aggregating signatures that are then gossiped to a proposer — a process that takes a full 32-slot epoch, and two epochs for finality.
They then model Wonderboom as a tree with fanout m, where every node is a committee of 128 validators with 16 random representatives, and where all N validators act as leaf validators in every slot. At the end of an epoch, a common random seed (e.g., Ethereum's RANDAO) is used to construct 32 distinct trees, so a different tree aggregates all N signatures in each of the 32 slots of the epoch.
They define four roles — leaf validators, leaf aggregators, internal aggregators, and the leader/proposer — and give explicit algorithms for each: leaf validators verify the previous aggregate against a bitvector-derived aggregate public key and then vote; leaf aggregators verify and aggregate up to m votes and forward the aggregate plus bitvector; internal aggregators collect aggregates from child committees, select the largest one from each, and aggregate upward; the proposer runs the same code but forwards the result into its block.
Censorship is tackled on two fronts. Proposer censorship is mitigated by physically separating proposer and validator processes onto distinct IP addresses. Vote censorship is mitigated by the largest-plus-one-random forwarding rule and the once-per-k reward window, which together yield a formal definition of α-vote-censorship resilience with α = 1 − (1−p)^k.
For performance, they derive aggregation time as a combination of computational cost (public key aggregation, signature aggregation, signature verification) and network delay Δ·d for tree depth d, and they analyze a two-part tree (depth 1 with fanout m, plus a tree of fanout m/16 and depth ⌈log_{m/16}(N/16)⌉, giving total depth ⌈log_{m/16}(N/16)⌉ + 1) to identify the fanout that minimizes aggregation time. They then build a simulator to test worst-case behavior at the million scale.
The system model assumes synchrony with a known bound Δ (necessary because slashing is used for accountability), an adversary corrupting up to f of N = 3f+1 nodes, a slowly adaptive adversary that can only change the corruption set at epoch boundaries, PKI-based authentication, and BLS signatures for aggregating N signatures and N public keys into a single verifiable pair.
Why This Matters
Impact on research. The paper reframes censorship analysis in Ethereum from committee composition to committee representation, and shows the standard metric overstates resilience by several orders of magnitude. It also formalizes vote-censorship resilience at scale — a property prior work treated informally — and it contributes the first simulation tool capable of evaluating large-scale Byzantine fault tolerant aggregation under worst-case N/3 faults. The result is a reusable analytical and experimental foundation for any protocol that must aggregate signatures from millions of participants under a fixed time budget.
Real-world applications:
- Payments and point-of-sale: Reducing finality from roughly 15 minutes to approximately 24 seconds makes on-chain settlement plausible for everyday transactions.
- Cross-chain bridges and exchanges: Faster, single-slot finality reduces the capital and time exposure of intermediaries that wait on Ethereum finality.
- Stablecoin savings in high-inflation economies: The paper notes millions of people rely on Ethereum to hold foreign-currency savings; reducing censorship-driven stake drift protects the safety margins those users depend on.
- Staking pools and validator operators: The once-per-k reward scheme and proposer/validator separation change the operational and reward model for entities running large validator fleets.
Industry relevance. The paper frames Ethereum as securing more than $650 billion in assets — placing it among the top 25 stock exchanges worldwide, ahead of Singapore, Mexico, and Thailand — with validators locking 32 ETH (over $96,000 at the time of writing). Because the design is modular and explicitly targets compatibility with existing Ethereum parameters (12-second slots, the 128/16 committee structure, RANDAO), it is positioned as an incremental upgrade path rather than a redesign.
Future Directions
-
Choosing and validating k: The paper explicitly defers discussion of the implications of different values of k to its later sections and flags free-riding — nodes participating only every k rounds because they lack incentive to participate each round — as a concern to be examined in more detail.
-
Deployment cost of proposer/validator separation: The authors state that Section 7 discusses how the two-process split can be set up "without significantly increasing the hosting costs or complexity for validators" — a practical question requiring operational validation.
-
Parameterizing the tree: The tree fanout m trades even computational load (smaller m) against network latency Δ·d (larger depth), and the paper's optimal-tree identification is derived under simplifying assumptions (e.g., aggregators include only the largest aggregate per child committee) that invite empirical follow-up.
-
Progress under δ < Δ: The authors note, citing prior work, that the protocol may still make progress under a smaller synchrony bound δ < Δ; characterizing this behavior at scale is left open.
-
Integration with production clients: Since Wonderboom is presented as a black box providing vote aggregation for any leader-based consensus protocol Ω, an obvious next step is measuring it inside a real client and network rather than in simulation.
Target Audience
This paper is most valuable to blockchain protocol researchers and engineers working on consensus layer design — particularly those involved in Ethereum client development, staking infrastructure, or signature-aggregation schemes (BLS, bitvector-based key aggregation). It also suits distributed-systems researchers interested in Byzantine fault tolerance at extreme scale, and security analysts studying censorship and MEV-driven attacks. Graduate students with background in BFT consensus and threshold/aggregate signatures will gain the most; readers without that background will find the model and algorithm sections dense but the motivation and economic analysis accessible.
Authors’ abstract
Over the last years, Ethereum has evolved into a public platform that safeguards the savings of hundreds of millions of people and secures more than…
The full abstract on arXiv