Foundations of Transaction Fee Mechanism Design — Hao Chung, Elaine Shi (2021)
Transaction Fee Mechanism Design — Tim Roughgarden (2021)
Transaction Fee Mechanism Design in a Post-MEV World — Maryam Bahrani, Pranav Garimidi, Tim Roughgarden (2023)
Maximizing Miner Revenue in Transaction Fee Mechanism Design — Hao Chung (2023)
Collusion-Resilience in Transaction Fee Mechanism Design — Hao Chung, Tim Roughgarden, Elaine Shi (2024)
Barriers to Collusion-resistant Transaction Fee Mechanisms - Yotam Gafnim, Aviv Yaish (2024)
A line of academic work, led by Elaine Shi (Cornell) and Tim Roughgarden (Columbia), has produced a series of papers arguing that incentive-compatible blockchains, or “transaction fee mechanisms,” cannot be designed under certain formal assumptions.
This page explains why these papers do not provide grounds for general claims about what is possible in distributed consensus. The issue is not the mathematical correctness of their theorems within their formal models; it is that the models themselves have three systemic problems that prevent their conclusions from holding even within their own restricted settings. The problems are:
off-equilibrium path exclusion: pruning strategically-relevant deviations through which mechanisms can enforce honest behavior via credible deterrence;
improper reduction: altering the strategic environment when invoking the Revelation Principle to produce a direct mechanism that is not strategically equivalent to the original game;
dimensional collapse: shifting between single-parameter and multi-dimensional assumptions when connecting underlying agent preferences to in-mechanism behavior.
When analyzing user strategies, the TFM papers assume miner honesty; when analyzing miner strategies, they assume user honesty. By cleaving the strategy space in this way, their model excludes mechanisms that deliver incentive compatibility or change preferred in-equilibrium outcomes through the threat of credible punishments in off-equilibrium paths.
The Revelation Principle states that if some outcome can be implemented by any mechanism, then there exists a direct mechanism equivalent. Myerson requires the direct mechanism to preserve the full strategic environment of the original game. The model must be able to implement all successful as well as unsuccessful equilibria. All preferences expressable in the full game must remain expressable in the reduced-form game.
The TFM papers fail this requirement, invoking the Revelation Principle in a manner that shrinks the strategy space rather than preserves it. Actions that are acknowledged to be feasible and strategically-relevant in the real mechanism -- such as the insertion of fake transactions by miners or side-contract payments from users to miners -- are removed entirely from the strategy space in the reduced mechanism.
This shift breaks the equivalence the Revelation Principle is meant to guarantee, and makes it impossible to generalize results back from the (non-)equivalent model to the original environment and full-form mechanism.
The TFM papers ask users to share their private valuations for “blockspace” with the mechanism. Yet additional forms of utility -- preferences for timing, transaction ordering, MEV protection, side-payments and other benefits of collusion -- are acknowledged as factors that affect strategic behavior.
This vacillation between acknowledging a multi-dimensional reality and suppressing it introduces internal contradictions. Classical single-parameter assumptions are needed to ensure monotonicity assumptions or apply Myerson’s Lemma. But unrevealed preferences are invoked to justify agent preferences for off-equilibrium deviations.
Although the TFM literature adopts the language of incentive compatibility, the way it models incentive compatibility and collusion-resilience is not aligned with the formal requirements of economics.
The differences break the conditions that Hurwicz, Maskin, and Myerson establish for studying the implementability of social choice rules. As a result, foundational results like the Revelation Principle cannot be invoked to generalize about implementability. What remains is simply an analysis of equilibrium behavior inside specific models with restricted message spaces.