Research
Price of Censorship: Censorship Resistance and Throughput under Rational Concurrent Proposers
Within a single block, the block proposer is a monopolist, so an adversary who wants to exclude a transaction only needs to bribe one party — and only needs to match the fee that transaction would…

- arXiv
- 2607.16995
- Published
- 2026-07-18
- Authors
- Aditya Saraf, Ioannis Kaklamanis, Sarisht Wadhwa, Fatima Elsheimy
AI summary
Overview
Research area: Blockchain economics and mechanism design (cs.GT) — specifically transaction fee mechanisms (TFMs), censorship resistance, and multi-proposer block construction.
Technical level: Intermediate. The paper is written for readers comfortable with basic game theory (Nash equilibrium, mixed strategies) and blockchain fee markets, but the core ideas and results are explained in accessible terms.
Scope: The paper builds a game-theoretic model of economically motivated censorship under multiple concurrent proposers (MCP), defines a quantitative metric of economic censorship resistance (eCR), and compares three transaction fee mechanisms on the censorship-resistance versus throughput trade-off.
What This Paper Is About
Within a single block, the block proposer is a monopolist, so an adversary who wants to exclude a transaction only needs to bribe one party — and only needs to match the fee that transaction would have paid. The paper asks how much better a system becomes when block-building is split among n concurrent proposers whose sub-blocks are merged, and how the answer depends on the fee mechanism that decides who gets paid when several proposers include the same transaction. The central tension is that redundancy protects transactions but also causes duplication, which consumes the throughput that parallelism was supposed to create.
Key Contributions
-
A quantitative metric for censorship cost. The authors define economic censorship resistance (eCR) as the adversary's expected total expenditure to censor a transaction divided by the user's expected payment for inclusion, together with formal definitions of fixed-mempool throughput, transaction cost, and bribery cost under a mempool drawn i.i.d. from a tip distribution F.
-
A general equilibrium algorithm. They characterize symmetric mixed Nash equilibria of the proposer game via a linear program and LP duality, showing that all transactions receiving interior inclusion probability share a common marginal payoff threshold λ*. This reduces equilibrium computation to a one-dimensional bisection that works for a broad class of "admissible" payoff functions — covering the duplication-penalizing, even-split, and collective TFMs with or without a bribe.
-
Closed-form bribery results. A "cut-and-paste" argument yields closed-form expressions for the bribe needed to censor any target transaction to any target probability, plus explicit inverses f_i(λ) for each of the three TFMs analyzed.
-
A TFM comparison with empirical grounding. Simulations on synthetic Pareto bid distributions vary the number of proposers, block congestion, the adversary's target censorship probability s (typically s = 0.9), and the amount of fee burn; findings are then checked against the empirical tip distribution of 10,000 recent Ethereum blocks (dataset released on Zenodo).
Main Findings
-
Single-proposer systems are structurally weak. Under a first-price auction the adversary need only match the user's bid, so eCR is at most 1. Fee burning makes this worse: on Ethereum the proposer receives only the tip while the base fee is burned, and with tips of roughly 0.5–2 gwei against base fees of 10–50 gwei, eCR ≈ tip/(base fee + tip) falls in [0.02, 0.1]. An adversary can censor by paying a few percent of what the user pays.
-
The duplication-penalizing TFM dominates. Across the settings tested, the TFM that pays a transaction's tip to its includer only if the transaction is included exactly once outperforms both the even-split TFM and the collective (pooled, equally divided) TFM on both eCR and throughput.
-
eCR scales linearly in the number of proposers. Because the adversary must bribe many proposers, censorship cost grows roughly proportionally with n, and this linear scaling survives fee burning.
-
Throughput plateaus quickly. Beyond the one-time duplication cost of moving away from a single proposer, throughput approaches the optimistic bound given by uncoordinated uniform sampling; adding more proposers buys censorship resistance at essentially no additional throughput cost.
-
Burning does not defeat MCP. At Ethereum's early-2026 burn levels, where the base fee is approximately 8.7 times the average tip, a single proposer yields eCR ≈ 0.09, whereas 49 concurrent proposers yield eCR ≈ 2.5 — meaning the adversary must outspend the user rather than pay a few percent of their fee.
-
Censorship is empirically persistent. The paper cites measurements that 46% of post-merge Ethereum blocks were produced by actors censoring OFAC-sanctioned transactions, and that as of early 2026 roughly 35% of Ethereum builders and half of all relays engage in some form of transaction censorship.
Methodology in Plain English
The authors set up a stylized game. A block of total size B is split among n proposers, each of whom independently selects up to B/n transactions from a shared mempool, and the final block is the union of their sub-blocks. An adversary publicly offers every proposer an unconditional bribe b for excluding one target transaction; the bribe is common knowledge. Proposers are rational, do not collude, build their sub-blocks simultaneously, and treat user tips as exogenous.
Because duplication means one proposer's payoff depends on what the others do, the solution concept is a symmetric mixed Nash equilibrium, summarized as a vector of per-transaction inclusion probabilities. The authors show that a proposer's best response to fixed beliefs is a linear program with a block-size budget, and that LP duality implies a threshold structure: transactions above a common marginal payoff λ* are always included, those below are never included, and the rest are included with interior probability. Since each transaction's equilibrium probability is pinned down by λ* alone, the equilibrium can be found by binary search on a single scalar — an efficient procedure valid for any admissible payoff function (one that is continuous, strictly decreasing in the competition probability, and hits the stated boundary values).
With this machinery, the authors compute expected throughput, expected user payments, and the minimum bribe achieving a target censorship probability, then aggregate over mempools drawn from a bid distribution. Simulations sweep the number of proposers, congestion, target censorship probability, and burn level on synthetic Pareto bids across the three TFMs. A final validation step feeds the empirical tip distribution from 10,000 Ethereum blocks into the same pipeline.
Why This Matters
Impact on research. The paper supplies a reusable framework — the eCR metric plus a general equilibrium solver — for reasoning about censorship as an economic cost rather than only as a worst-case liveness property. It complements Byzantine-fault-tolerance analyses such as those on FOCIL and related inclusion-list designs, which bound worst-case delay imposed by faulty parties, by instead pricing deviations by rational, economically motivated parties. It also connects directly to prior work on multi-proposer fee mechanisms (Fox et al.'s two-tier tips, Stouka et al.'s "Double" and "Single" TFMs, Garimidi et al.'s FPA-EQ, and AUCIL's use of the even-split TFM, which the paper analyzes as one of its admissible designs).
Real-world applications:
- Ethereum and Solana multi-proposer proposals — the analysis speaks to active designs that split block production across concurrent proposers, and identifies which fee-splitting rule to pair with them.
- DAG-based and leaderless chains such as Sui (Mysticeti) and Aptos, and research protocols such as MirBFT, which already allow many validators to propose blocks simultaneously.
- MEV-sensitive applications — on-chain auctions, DeFi liquidations, and arbitrage races, where excluding a competitor for a few blocks is worth orders of magnitude more than the fees at stake, making censorship profitable.
- Bribery resilience beyond censorship — the same reasoning about per-block censorship cost connects to work on bribery against optimistic-rollup fraud proofs, where the analysis assumes a per-block cost that eCR is designed to quantify.
Industry relevance. With roughly 35% of Ethereum builders and half of all relays reported to filter transactions, and about 46% of post-merge blocks produced by actors censoring OFAC-sanctioned transactions, the finding that a well-designed TFM plus more proposers can lift eCR from about 0.09 to about 2.5 is directly actionable for protocol designers. The authors note that MCP is a "weaker but more flexible" tool than inclusion lists: proposers in restrictive jurisdictions can still censor if legally required, which avoids pushing stake into fewer geographic regions, while the transaction becomes more lucrative for proposers elsewhere.
Future Directions
- Conditional bribes. The threat model uses unconditional bribes. Adversaries who pay only upon successful censorship spend less, so extending the equilibrium computation to conditional bribes is a natural next step and would give a tighter (lower) eCR.
- Endogenous user bidding. Tips are treated as exogenous and user–proposer collusion is not modeled. Under the duplication-penalizing TFM, duplicated transactions execute free of charge, and a user coordinating with even one proposer could deliberately trigger duplication to avoid paying; user-side equilibrium analysis is left open.
- Intermediate two-tier TFMs. The paper analyzes the extreme version of Fox et al.'s two-tier design. Intermediate versions that charge a small tip on duplicated inclusion are admissible under the paper's definition and should be evaluated, since they may mitigate both the revenue loss and the collusion concern.
- Private mempools and private order flow. The centralized role of private order flow motivates a private-mempool extension discussed but not modeled; it would change which transactions each proposer can see and therefore the duplication equilibrium.
- Multi-block and repeated censorship. The analysis covers censoring a transaction within a single block; repeated censorship across many sequential blocks, and the richer adversary models that come with it, remain outside the current model.
Target Audience
Protocol designers and mechanism-design researchers working on fee markets, proposer–builder separation, or multi-proposer consensus will get the most direct value. It is also useful for economists studying bribery and censorship in decentralized systems, for blockchain engineers evaluating MCP proposals for Ethereum or Solana, and for graduate students in algorithmic game theory who want a worked example of LP duality applied to a concrete equilibrium computation. Readers need some comfort with mixed-strategy equilibrium concepts, but the intuition behind each result is stated in plain terms.
Authors’ abstract
Censorship resistance is the defining advantage of blockchains over their centralized counterparts. Yet block proposers censor transactions for many reasons, from legal consequences to economic incentives. We study economically-incentivized censorship, modeled by an adversary who bribes proposers to exclude a target transaction, and define the economic censorship resistance (eCR) of a transaction as the adversary's expected cost of successful censorship divided by the user's…
The full abstract on arXiv