Skip to content
Crypto.world

Research

One Node, Two Roles: Simultaneous Contests for Validation and Attention in Rollups

Optimistic rollups are safe if at least one honest validator executes the state transition function and disputes bad assertions, but verifying is unpaid whenever every assertion is correct, so few…

One Node, Two Roles: Simultaneous Contests for Validation and Attention in Rollups
arXiv
2610.12220
Published
2026-10-08
Authors
Pranay Anchuri, Ben Berger, Matteo Campanelli, Akaki Mamageishvili

AI summary

Overview

Research area: Rollup economics and mechanism design, combined with cryptographic proof systems (succinct arguments, inner-product arguments, knowledge soundness).

Technical level: Advanced. The paper assumes familiarity with optimistic rollups, dispute protocols, interactive/succinct proof systems, and contest-theoretic equilibrium analysis.

Scope: The paper builds a joint economic and cryptographic framework for why rollup "attention" rewards may buy registered identities rather than independent executions, and uses it to compare two attention mechanisms (TRACE and Proof of Diligence), with a worked case study on Arbitrum One.

What This Paper Is About

Optimistic rollups are safe if at least one honest validator executes the state transition function and disputes bad assertions, but verifying is unpaid whenever every assertion is correct, so few parties are willing to do it. "Attention" mechanisms pay nodes to execute regardless of validation, yet a single operator can back many registered identities with one execution, so paying per identity does not necessarily produce independent execution. The paper models validation and attention as two simultaneous, coupled contests sharing one execution cost, and asks what per-identity costs are needed to sustain a target number of genuinely independent operators.

Key Contributions

  1. A two-contest game-theoretic model. Validation and attention are analyzed as two coupled Tullock-like contests in which an operator pays a one-time node cost c, pays a marginal cost d_a per attention sybil, and each role's reward is split in proportion to registered identities. The authors derive conditions on entry and marginal sybil costs that support a target execution diversity in equilibrium.

  2. A new cryptographic primitive: arguments with registered provers. The paper formalizes completeness, binding, knowledge soundness, succinctness, designated proving, and a new property called non-amortizability with overhead Δ, meaning that producing valid proofs under n distinct keys takes time roughly n·Δ. The authors state that non-amortizability of this kind has not been studied for modern succinct proofs before.

  3. A cryptographic comparison of TRACE and Proof of Diligence. TRACE is recast as an argument with registered provers with Δ = Ω(log N · λ^{1-ε}) (Theorems 3.9 and 1), while Proof of Diligence has Δ = O(1). This difference means the per-identity cost in TRACE can be tuned via trace length, whereas in Proof of Diligence it is a constant the deployer cannot raise.

  4. A worked case study on Arbitrum One plus a stronger sketch construction. Measurements give d_a = $0.91 per hour at trace length N = 2^20; non-decreasing rewards then support at most A = 2 independent operators, while N = 2^24 raises this to A = 29, and decreasing reward functions reach A = 30 at the baseline cost. An Appendix I construction (registered PRF key plus block, commit-and-prove) achieves overhead linear in trace length, at the price of a more expensive prover.

Main Findings

  • The safety assumption is cheap to state, costly to satisfy. The 1-of-N safety guarantee of optimistic rollups requires someone willing to execute and to lock stake; under Arbitrum's BoLD dispute protocol the stake is at least 3600 ETH [Arb24], and validators are typically uncompensated while all assertions are correct. The paper frames this as the verifier's dilemma of Luu et al. [LTKS15] in rollup form.

  • Paying per identity does not buy execution diversity. Execution diversity A is defined as the number of distinct operators independently executing the STF, not the number of registered identities. A single operator can register many sybils sharing one execution trace, so rewards can buy claims rather than executions.

  • A concrete equilibrium bound for non-decreasing rewards. When the total reward does not decrease as identities are added, one identity per operator is an equilibrium only if A ≤ 1 + 2d_a/c. The larger the per-identity cost relative to the node cost, the more independent operators the reward can sustain; when an extra identity is cheap, an incumbent captures more reward by adding an identity than a newcomer can by running a node.

  • Decreasing reward functions are an alternative lever. A reward function that decreases past some threshold avoids the identity-proliferation problem, whereas under a non-decreasing reward d_a itself must be adjusted to support more independent attention operators in equilibrium.

  • TRACE's non-amortizability comes from its inner-product argument. TRACE uses a Bulletproofs-like inner-product argument [BCC+16, BBB+17], instantiated with the publicly verifiable variant of Daza et al. [DRZ20], whose prover runs log N rounds that fold committed vectors using Fiat–Shamir challenges. Because the public key seeds these challenges, each folding step yields a key-dependent combination of the entire trace that no other identity can reuse, giving Δ = Ω(log N · λ^{1-ε}).

  • The guarantee is weaker than the honest prover's cost. Δ grows only logarithmically in N, while the honest prover does work linear in N.

  • Proof of Diligence has constant per-identity overhead. Each identity evaluates a VRF on the Merkle roots of the new state and of the execution trace and is paid if the output falls below a stake-dependent threshold. The proof certifies the roots, not the trace, and its input is shared by every identity of an operator, so Δ = O(1) and the equilibrium bound on A stays fixed under a non-decreasing reward.

  • Proof of Diligence does not guarantee trace knowledge. A node that obtains the correct root r_E from another node can produce an accepting proof without executing. The mechanism relies on an incentive-based approach: watchtowers recompute r_E and challenge incorrect proofs during a watchtower challenge period, and selling a computed r_E is treated as collusion; with a whistleblower contract, independent diligence is the unique equilibrium for rational and independent watchtowers [SRB+24, Cor. 8]. That guarantee is stated not to apply to an operator that computes r_E once and reuses it for all its own identities.

  • TRACE claims are knowledge sound in the paper's sense. A mechanism whose proofs are knowledge sound in the sense of Definition 3.6, as TRACE's are [Anc26], rules out producing an accepting proof without the trace, since an extractor can recover the trace from an accepting proof.

  • Arbitrum One numbers. At d_a = $0.91 per hour and trace length N = 2^20, non-decreasing reward functions support execution diversity of at most A = 2. Raising trace length to N = 2^24 raises the maximum to A = 29; reward functions that decrease beyond a threshold support A = 30 at the baseline cost.

  • A stronger but costlier construction is sketched. Appendix I derives a secret vector from a registered PRF key and the block, so every identity folds the trace against fresh pseudorandom values each block, with a general-purpose commit-and-prove argument showing the vector was derived correctly. This raises overhead to linear in trace length, within a logarithmic factor of the honest prover's cost, and treats the PRF as a random oracle because the adversary in Definition 3.7 holds every key.

Methodology in Plain English

The authors start from an observation about incentives: in an optimistic rollup, a node that has already executed a batch holds the execution trace, and that single execution can support any number of registered identities. They therefore split an operator's costs into a single node cost c, paid once for execution, and a per-identity cost d_a, paid for each additional registered identity in the attention role.

They then model operators as risk-neutral players in two simultaneous contests — one for validation rewards, one for attention rewards — coupled through the shared execution cost. Working with expected payouts rather than individual lottery draws, each sybil earns an equal expected share of its role's reward, which makes the game a pair of Tullock-like contests. Solving for equilibria yields conditions relating c, d_a, and the achievable execution diversity A.

To say something rigorous about d_a, the authors need a cryptographic statement about how much work each additional identity must do. They introduce "arguments with registered provers," in which each identity has a key pair and must produce a proof under its own key about a committed execution trace, and define non-amortizability: an adversary holding n keys cannot produce n valid proofs in much less than n·Δ time. They then instantiate this definition for TRACE, proving the overhead bound by counting group operations round by round in the generic bilinear group model (Appendix A), and compare it with Proof of Diligence's VRF-over-roots approach, which has constant overhead. Finally, they plug measured costs for Arbitrum One into the equilibrium bound to get concrete diversity limits.

Why This Matters

Impact on research. The paper argues that attention and validation have been studied separately even though they share the same sunk execution cost, and it supplies the first joint analysis the authors are aware of. It also introduces non-amortizability as a property of succinct arguments over non-deterministic computations, contrasting with prior non-amortizability work in proofs of useful work [BRSV17, KW25], and gives a formal language for comparing attention mechanisms rather than describing them informally.

Real-world applications.

  • Designing reward schedules for rollup attention/validator programs so that payouts attract independent operators rather than sybil identities.
  • Choosing the trace length parameter N as an explicit policy lever: on Arbitrum One, moving from 2^20 to 2^24 changes the attainable diversity bound from A = 2 to A = 29.
  • Supplying block data to light clients and wallets through a larger pool of attention nodes that actually hold execution data [ABSBK21].
  • Evaluating whether a proposed attention mechanism (TRACE-style proofs versus VRF-over-roots designs such as Proof of Diligence) provides cryptographic or merely incentive-based assurances about trace knowledge.

Industry relevance. The authors are at Offchain Labs, and the case study targets Arbitrum One, a widely deployed optimistic rollup, making the cost figures and parameter trade-offs directly relevant to live deployment decisions rather than purely theoretical settings.

Future Directions

  • Costing the stronger construction. The paper explicitly leaves a concrete cost estimate for the Appendix I construction to future work, which is needed before its linear-overhead guarantee can be weighed against its more expensive prover.

  • Preventing trace resale cryptographically. The paper notes that neither mechanism rules out an operator obtaining the trace from another; TRACE argues buying the trace is unprofitable once bandwidth cost exceeds expected reward, and Proof of Diligence treats resale as collusion. A cryptographic rather than incentive-based solution remains open.

  • Extending the game beyond expected payouts. The analysis does not model individual lottery draws and works with expected payouts; richer models (for example, incorporating risk attitudes or the variance of lottery outcomes) are a natural extension.

  • Broadening the empirical base. The quantitative results come from Arbitrum One measurements at N = 2^20 and 2^24 with d_a = $0.91 per hour; measuring c and d_a across other rollups and tracing the full cost/reward trade-off of longer traces would test how general the bounds are.

Target Audience

Researchers and protocol designers working on rollup incentivization, mechanism design for blockchains, and applied cryptography for proof systems. It is most useful to readers who already understand optimistic rollup dispute protocols and are comfortable with both equilibrium analysis and succinct-argument security definitions; newcomers can read the Overview and Arbitrum One case study for the practical conclusions but will need background for Sections 3 and 4.

Authors’ abstract

An optimistic rollup is safe as long as at least one honest validator executes its state transition function (STF) and disputes any assertion that is inconsistent with the result of the execution. Allowing anyone to participate, however, does not give incentive to do so: as long as every assertion is correct, verifying assertion correctness goes unpaid. Attention mechanisms address this by paying nodes to execute the rollup's STF, regardless of whether they participate in…

The full abstract on arXiv
Read the original paper