Skip to content
Crypto.world

Research

From PBS to ePBS: the Microstructure of Block Building

Ethereum's current PBS implementation, MEV-Boost, routes builder bids through relays, and by July 2026 the top five relays accounted for over 90% of block production.

From PBS to ePBS: the Microstructure of Block Building
arXiv
2607.11240
Published
2026-07-13
Authors
Jingyu Liu, Bolin Zhang, Lin William Cong, Siguang Li, Xuechao Wang

AI summary

Overview

Research area: Blockchain economics and market microstructure, specifically auction and mechanism design applied to Ethereum block building (cs.GT; JEL D44, D47, G10, G23, L14).

Technical level: Advanced. The paper relies on perfect Bayesian equilibrium (PBE), information design, Myerson optimal auction theory, and counterfactual regret minimization.

Scope (one sentence): The paper models how Ethereum's shift from relay-mediated proposer-builder separation (PBS) to enshrined PBS (ePBS) changes bidding incentives, latency rents, and proposer revenue, and proposes a Trusted Execution Environment (TEE) sidecar to recover part of the lost commitment power.

What This Paper Is About

Ethereum's current PBS implementation, MEV-Boost, routes builder bids through relays, and by July 2026 the top five relays accounted for over 90% of block production. The upcoming Glamsterdam upgrade introduces ePBS, which removes the relay layer and lets proposers receive signed builder bids directly, giving them control over when to propose a block and what bid information to disclose. The paper asks how that newly granted flexibility — deferring the proposal and selectively revealing early bids — reshapes builder competition, allocation efficiency, and proposer revenue relative to the PBS benchmark.

Key Contributions

  1. A unified two-stage block-building game. The authors define a general imperfect-information, two-stage auction with verifiable messages in which an early bid is simultaneously a price offer and a signal. PBS and ePBS are then derived as two restrictions of the same game: PBS fixes the stopping profile and disclosure exogenously (via a parameter q₁), while ePBS makes the proposer's stopping and disclosure choices sequentially rational best responses.

  2. Latency modeled as asymmetric information updating. A builder is "fast" if the proposer-to-builder-to-proposer round-trip time is short enough for the continuation message and a revised stage-2 bid to return before the second-stage stop; otherwise it is "slow." Fast builders observe whether the auction continued and any public and private verifiable messages before rebidding, while slow builders do not update at all.

  3. Exact equilibrium characterizations plus calibrated no-regret computation. The paper proves PBE results in tractable cases — an all-slow first-price-auction (FPA) benchmark, a selection result for PBS, immediate-stop thresholds for ePBS, and a family of uni-pooling ePBS equilibria — and cross-validates them with calibrated Counterfactual Regret Minimization in finite games with non-i.i.d. values, heterogeneous latency, and private disclosure.

  4. A commitment-failure diagnosis and a TEE-based mitigation. The ePBS distortion is interpreted as a commitment failure whose full-commitment limit is the static Myerson auction. The authors propose a TEE sidecar that enforces limited commitment over stopping and disclosure while preserving the highest-bid, pay-as-bid termination rule, formulate the revenue-maximizing policy as a polynomial-size bilinear program, and report roughly a 25% proposer revenue gain over the first-price benchmark in conservative finite benchmarks.

Main Findings

  • PBS preserves the standard FPA benchmark under separation. Any PBE with separating stage-1 bids delivers the same interim builder payoff as the standard first-price auction, and in the regular pure all-slow class every such equilibrium is diagonal — x(v) = z(v) = β^FPA(v) for F-almost every v — hence payoff-equivalent to the FPA.

  • Existence of symmetric separating PBS equilibria depends on the stage-1 settlement probability. Under i.i.d. regular values with n builders, a symmetric strictly separating PBS PBE exists whenever q₁ ≥ 1/2, whereas no such equilibrium exists when q₁ < 1/n. For i.i.d. uniform values the nonexistence region strengthens to q₁ ≤ 1/3.

  • An all-slow benchmark collapses both institutions to the one-shot FPA. If all builders are slow, the diagonal complete plan (b_{i,1}, b_{i,2}) = (y_i, y_i) with y_i drawn from any FPA equilibrium bid rule can be completed into a PBE of any PBS game with q₁ in (0,1) and of any ePBS game, where the proposer stops with probability one at every on-path stage-1 information set.

  • ePBS has a stopping frontier k₂*. Under a non-overbidding constraint, if k₂ is at or below the lower threshold k̲₂(β₁), stopping is sequentially optimal at every reached stage-1 history regardless of latency profile; if all builders are fast and k₂ exceeds the upper threshold k̄₂(β₁), some reached history makes continuation with full public disclosure strictly better. For a pure symmetric strictly increasing efficient FPA bid rule the two thresholds collapse to k₂*(β^FPA) = inf over x > 0 of k₁ β^FPA(x)/x; in the i.i.d. uniform n-builder case with k₁ = 1 this equals (n−1)/n.

  • A ratchet effect distorts ePBS once continuation is credible. Because the proposer can defer the proposal and reuse early bid information in stage 2, builders anticipate ex-post extraction and shade or pool their early bids. In a tractable two-fast-builder ePBS environment the authors construct a family of uni-pooling PBEs in which low types pool at zero and high types separate on a positive branch, and feasible selections generate both lower proposer revenue and lower allocation efficiency than the FPA benchmark.

  • The distortion appears as a revenue-efficiency valley in calibrated games. Calibrated to real-world builder valuations, PBS approaches FPA-like outcomes as stage-1 settlement becomes more likely, while ePBS develops a revenue-efficiency valley once continuation becomes credible. The calibrated results also show that ePBS compresses, but does not eliminate, the fast-builder latency premium.

  • Full commitment removes the ratchet channel. An institutional proposer with full ex-ante commitment implements an optimal committed proposer rule that coincides with the Myerson auction outcome, eliminating the ratchet-induced revenue distortion.

  • Limited commitment via TEE recovers part of the gain. With the TEE enforcing an ex-ante stopping and disclosure policy while keeping the terminal highest-bid, pay-as-bid rule, the design problem becomes a constrained information-design problem: which TEE policy induces a PBE with maximal proposer revenue. In a discrete benchmark the optimal TEE policy raises proposer revenue by roughly 25% relative to FPA across latency profiles.

  • Institutional context documented. Ethereum mainnet under current PBS has a proposal-success rate of about 99.69%; proposers' timing is not generally known to builders; builders may submit thousands of bids to relays within a 12-second slot; and deferred proposal reduces the probability that the selected bid becomes canonical, with a block counted as canonical when at least two thirds of attesters attest to it. The paper notes Titan, the dominant builder, expects peer-to-peer bids to be used sparingly because they are much slower than direct bids.

Methodology in Plain English

The authors start from the institutions: they describe the four-step per-slot pipeline (MEV identification, block building, block auction, proposal and attestation) and contrast the relay-mediated PBS flow with the protocol-facing ePBS flow. They use public MEV-Boost winning-bid timing data, real-world geolocations, and Google Cloud latency data to simulate block propagation and translate network delay into region-specific proposal reliability, which supplies the reduced-form canonicalization probability used in the model.

They then build a two-stage game. Builders learn private values and submit stage-1 bids; the signed bids form a verifiable history the proposer cannot fabricate. In ePBS the proposer decides whether to propose immediately or defer and send public and private verifiable messages; fast builders see the continuation event and messages in time to respond, slow builders do not. A stage-2 bid cannot go below the builder's own stage-1 bid, and deferred blocks succeed with weakly lower probability.

Two solution methods are combined. Where the game is tractable, the authors prove PBE results directly — characterizing equilibria, checking sequential rationality of stopping, and deriving existence thresholds. Because realistic settings with non-i.i.d. values and private disclosure are analytically intractable, they also compute calibrated no-regret benchmarks with Counterfactual Regret Minimization. The authors are explicit that no-regret dynamics in general-sum extensive-form games approximate extensive-form coarse correlated equilibrium rather than exact PBE, so the computation is presented as a robustness and external-validity check rather than a substitute for the theory. Finally, the TEE design problem is written as a direct information kernel, its PBE feasibility constraints are characterized, and the design problem is reduced to a polynomial-size bilinear program solved in benchmark settings.

Why This Matters

Impact on research. The paper reframes ePBS not as a plumbing change but as a change in the auction's information structure: early bids become verifiable signals the proposer can exploit, which is a limited-commitment information-design problem rather than a standard auction. It also connects the block-building literature to credible auctions, sequentially optimal mechanisms, and persuasion under weak institutions, and it supplies a tractable template for combining exact PBE characterizations with calibrated no-regret computation.

Real-world applications:

  • Protocol design for Ethereum's Glamsterdam upgrade: identifying that proposer-controlled stopping and disclosure creates a ratchet distortion, and quantifying the canonicalization-probability frontier below which immediate proposal remains optimal.
  • Mitigation mechanisms: a TEE sidecar that commits the proposer ex ante to stopping and disclosure policies, evaluated as a revenue-improving, protocol-facing alternative to relays.
  • Relay and builder market structure: understanding how the 90%-plus concentration among the top five relays and the latency premium for fast builders might change once bids flow directly to proposers.
  • Proposer economics: distinguishing solo proposers with weak commitment technology from institutional proposers or relays that have stronger commitment through repeated interaction, reputation, or dedicated infrastructure.

Industry relevance. Builders, relays, proposer service providers, and protocol researchers all operate inside the trade-off the paper formalizes: waiting for a better bid improves the selected price but lowers the chance the block becomes canonical. The result that ePBS compresses but does not eliminate the latency premium is directly relevant to infrastructure investment decisions, and the roughly 25% revenue figure gives a concrete benchmark for how much value a credible commitment device could be worth to proposers.

Future Directions

  • Extending the equilibrium characterization beyond tractable cases. The uni-pooling analysis is carried out in a two-fast-builder environment; whether the ratchet channel produces qualitatively similar equilibria with more builders, mixed latency profiles, and richer disclosure sets remains open.
  • Making the TEE mitigation practical. The paper reduces the design problem to a bilinear program and evaluates it in a discrete benchmark; implementing and stress-testing such a sidecar against real builder behavior, adversarial proposers, and TEE trust assumptions is a natural next step.
  • Treating latency as a design variable rather than a constraint. Since fast versus slow is defined by the proposer-builder-proposer round-trip time relative to the stage-2 stop, the interaction between protocol timing parameters, geographic distribution, and equilibrium outcomes deserves a systematic treatment.
  • Connecting to competing ePBS proposals. The paper positions itself against concurrent work on relay versus in-protocol auctions, sealed first-price protocol channels, last-look disclosure, and non-leakage commitments; a unified comparison of these mechanisms under the same dynamic information structure would clarify which design best preserves allocative efficiency.

Target Audience

Protocol researchers and mechanism designers working on Ethereum block building; academic economists and computer scientists studying auctions with limited commitment and information design; relay operators, builders, and proposer service providers who need to anticipate how direct bids change bidding incentives; and graduate students in blockchain economics or algorithmic game theory looking for a worked example that pairs exact perfect Bayesian equilibrium analysis with calibrated no-regret computation.

Authors’ abstract

Ethereum's Glamsterdam upgrade introduces enshrined proposer-builder separation (ePBS), replacing relay-centric PBS with direct builder bids to proposers. We study how this shift changes the block-building microstructure through a general imperfect-information two-stage auction with verifiable messages, where an early bid serves as both a price offer and a signal. PBS and ePBS are modeled as restrictions of the same block-building game: PBS fixes stopping and disclosure…

The full abstract on arXiv
Read the original paper