# Timing Games in Responsive Consensus Protocols

Crypto.world

> Optimistically responsive consensus protocols are designed to run at the speed of the network rather than waiting for a worst-case delay. But validators earn MEV, and MEV tends to grow the longer a…

**arXiv**: 2510.25144

**Published**: 2025-10-29

**Authors**: Kaya Alpturer, Kushal Babel, Aditya Saraf

## AI summary

Timing Games in Responsive Consensus Protocols

## Overview

- **Research area:** Computer Science > Game Theory (cs.GT), applied to Proof-of-Stake consensus protocol design, MEV, and validator incentive mechanism design.
- **Technical level:** Advanced. The paper uses order statistics of latency distributions, infinitely repeated games, subgame-perfect Bayes-Nash equilibrium, and coalition-deviation (coalition-proofness) analysis.
- **Scope:** The paper builds a game-theoretic model of validators who profit from delaying block proposals in optimistically responsive consensus protocols, and shows that a time-decreasing block reward measured through a "timeliness vote" can make prompt proposing and honest reporting an equilibrium.

## What This Paper Is About

Optimistically responsive consensus protocols are designed to run at the speed of the network rather than waiting for a worst-case delay. But validators earn MEV, and MEV tends to grow the longer a block is held back, so a selfish leader is tempted to delay its proposal (a "timing game"), which destroys the very responsiveness the protocol was designed for. The paper asks whether responsiveness is still achievable when validators are rational and delay is profitable, and answers by designing a reward rule that pays less for slow blocks and by measuring block times through votes from other validators.

## Key Contributions

1. **A timing-game model for responsive BFT protocols that includes random pairwise latencies.** Unlike prior timing-game work, realized peer-to-peer latencies are hidden, so validators cannot distinguish an intentionally delayed proposal from an honestly early proposal that simply arrived late. Each of the n players is one unit of stake, a leader is sampled uniformly at random each round, and the quorum threshold is c = ⌊2n/3⌋ + 1 for most protocols.
2. **A "timeliness vote" mechanism for measuring proposal duration without trusted clocks.** Each validator reports the round duration it observed, and the protocol takes the m-th smallest report (an order statistic) as the measured duration. Block timestamps alone are rejected as manipulable. Under the assumption m > n − c, the m-th smallest vote always comes from a validator who contributed to the quorum, so the measurement is at least the true delay.
3. **A proof that time-decreasing rewards make early proposing plus honest voting a subgame-perfect Bayes-Nash equilibrium.** The required condition is that M(x) + B(x) is decreasing in x, with a linearly decreasing block reward B(x) = b₀ − bx that reaches zero at the timeout τ (so b₀ = bτ). The paper calls this a prisoner's dilemma structure in which mutual cooperation (proposing promptly) beats mutual defection.
4. **Coalition-resistance and fairness analyses.** The paper classifies coalitions as small (< k), medium ([k, m)), and large (≥ m) where k = m + c − n, and analyzes whether whales (single entities with co-located stake) or cooperating groups can profitably deviate. It then asks whether latency-dependent rewards widen the gap between well-connected and poorly-connected validators.

## Main Findings

- **Timing games are a prisoner's dilemma.** Proposing early speeds up the system and raises the rate of block reward for everyone; defecting by delaying captures some MEV but slows the system. In the absence of coordination, delaying is a dominant strategy for an individual, even though all validators would be better off if everyone proposed promptly. MEV therefore hurts validators in the long run by reducing the rate at which block rewards accrue.
- **Early-proposing truthful equilibrium exists.** Theorem 4.1 states that if ⟨M, B⟩ are time-decreasing and m > n − c, then the profile where every leader sets delay Δ = 0 and every voter reports its observed duration honestly is a subgame-perfect Bayes-Nash equilibrium. The paper notes the honest-voting half is easy in this model because a validator's vote affects other leaders' rewards, not its own.
- **Bounded-stake whales do not deviate.** A single entity controlling many units of stake is shown not to deviate as long as it is sufficiently small — for example, controlling less than 1/3 of total stake under standard parameter choices (the c = m = 2n/3 case, where the small-coalition cutoff is n/3).
- **Small coalitions still propose early.** For whales of size at most k, the optimal strategy is the honest early-proposing strategy (Proposition 5.1). For rational coalitions of size at most k, the optimal strategy is to vote 0 on each other's proposals while still proposing at Δ = 0 (Theorem 5.3) — they exploit the vote, but not the delay.
- **Large zero-voting coalitions are unstable under low latency.** Under a "z-low-latency" condition requiring z ≤ (n − c)/(3n − c), the paper shows that |C| − k validators inside a large coalition would prefer to defect back to the early-proposing truthful strategy, so the coalition would shrink below the large threshold. The intuition is that enough validators prefer the all-cooperate outcome to the all-defect outcome.
- **Small and medium rational coalitions are argued to be unstable informally.** If a validator can accumulate m "free" votes it would delay with impunity, and other validators have an incentive to join additional zero-voting pacts to reduce their own latency penalty, so membership is hard to predict and enforce.
- **Fairness costs are real but small.** Under dynamic (time-decreasing) block rewards, utilities become more sensitive to latency, worsening the gap between best- and worst-connected validators. The paper shows this theoretically in two latency models (nodes on a line, and two clusters) and in simulations on real-world latency data, concluding the degradation is marginal. (Specific simulation parameters and numerical fairness values are not reported in the provided excerpt.)
- **Time-rotating leaders do not solve MEV-driven delay.** The paper argues that a leader elected for a period rather than a round would simply withhold DeFi transactions until its final block, since CEX–DEX arbitrage profit over a time range T grows proportionally to T^(3/2) (citing MMR24).
- **Reducing leader election probability is a worse lever than reducing block reward.** Lowering a slow leader's chance of being elected makes honest voting harder to incentivize, because reducing someone else's election probability directly raises one's own.

## Methodology in Plain English

The authors set up a repeated game in which each round has a randomly chosen leader. The leader learns (through an "optimal-delaying oracle" that sees all latency realizations) the largest delay it can take without triggering a timeout, and then picks a delay. Blocks and votes propagate with random, hidden pairwise latencies, so a validator seeing a block cannot tell whether it was late because the leader dawdled or because the network was slow. Each validator votes on the block and appends the duration it observed; the protocol sorts those reports and uses the m-th smallest. The leader's payoff combines an increasing MEV term M(Δ + S) — where S is the time until the leader could have started proposing — with a block reward B of the measured duration. Utility is the long-run time-averaged reward rate, so slower rounds directly dilute everyone's earnings. Because delays also stretch rounds, the reward-rate denominator couples all validators together, which is what creates the prisoner's dilemma. The authors then check whether anyone — an individual, a whale, or a coalition of a given size — can profit by deviating, and separately compute fairness by comparing reward spreads across validators under static versus dynamic block rewards in analytic latency models and in simulations built on measured network latencies.

## Why This Matters

- **Impact on research:** It reframes responsiveness as an incentive problem rather than purely a protocol-design problem, and shows responsiveness can itself be used to discipline proposers. It also adds hidden, random latencies to the timing-game literature, which previously assumed delays were observable.
- **Real-world applications:** - MEV-aware validator reward design for Proof-of-Stake chains, including replacing static per-block rewards with duration-dependent ones. - On-chain or in-protocol measurement of block timeliness using committee votes, avoiding manipulable timestamps. - Understanding and mitigating observed timing games on Ethereum (where blocks are 12 seconds regardless of propagation) and Solana. - Assessing decentralization risk from validator geographic collocation, since better-connected validators earn more under latency-sensitive rewards.
- **Industry relevance:** Consensus teams and staking operators get a concrete, parameterized mechanism (choose m, c, τ, b) plus a warning that block rewards set too low make timing games inevitable — an observation the paper compares to prior work linking low Bitcoin block rewards to forking attacks.

## Future Directions

- **The meta-game of coalition formation.** The authors explicitly defer coalition formation dynamics to future work, since their instability argument for small and medium coalitions is informal and assumes coalitions can neither detect defectors nor identify members of other coalitions.
- **Medium and large whales.** The paper states it has no counter-argument for medium or large whales, only the observation that such an entity would have to control a very large share of stake and would likely violate the underlying BFT assumptions.
- **Practical parameter guidance and calibration.** Section 7 promises guidance on setting protocol parameters, but the concrete recommended values are not included in the provided excerpt.
- **Extending the measurement to real deployment.** The suggested implementation — timing votes submitted directly on-chain and aggregated over a window spanning several leaders — raises open questions about aggregation windows, strategic reporting over time, and interaction with inflation, which the paper defers to an appendix.
- **Richer latency and reward models.** The excerpt notes that block propagation and voting latencies are treated as one distribution; using separate distributions and more elaborate MEV growth functions is left open.

## Target Audience

Protocol researchers and mechanism designers working on Proof-of-Stake consensus, MEV, and validator incentives; game theorists interested in blockchain applications of repeated games and coalition-proofness; and engineering or economics teams at staking providers, L1 foundations, and MEV-related infrastructure who need to reason about block timing rewards. Readers should be comfortable with equilibrium concepts and order statistics; the paper is not introductory.

## Authors’ abstract

Optimistic responsiveness -- the ability of a consensus protocol to operate at the speed of the network -- is widely used in consensus protocol design to optimize latency and throughput. However, blockchain applications incentivize validators to play timing games by strategically delaying their proposals, since increased block time correlates with greater rewards. Consequently, it may appear that responsiveness (even under optimistic conditions) is impossible in blockchain…

[The full abstract on arXiv](https://arxiv.org/abs/2510.25144)

[Read the original paper](https://arxiv.org/abs/2510.25144)

---

Canonical: https://crypto.world/paper/timing-games-in-responsive-consensus-protocols-2510.25144
