Auction-Based Backrunning Mechanisms

• Auction-based backrunning is a mechanism that allocates profitable post-trade sequencing rights via explicit and implicit auctions, including block-level MEV-Boost and timing contests.
• It examines how latency, strategic bidding (naive, adaptive, last-minute, bluff), and private orderflow asymmetry affect builders' valuations and winning chances.
• The literature employs continuous-time models, Stackelberg resilience analysis, and probabilistic timing games to assess auction efficiency, revenue impact, and protocol design implications.

Auction-based backrunning denotes the allocation of profitable post-trade sequencing rights through explicit or implicit competition over inclusion and ordering. In the Ethereum-centered literature, the concept spans multiple layers rather than a single canonical mechanism: builders compete in the MEV-Boost auction for block-construction rights under proposer-builder separation; agents can manipulate transaction fee auctions through credible on-chain commitments; and first-come-first-served ordering can induce an implicit all-pay contest in which repeated transaction submissions function as costly bids for the right to land immediately after a triggering event (Wu et al., 2023, Landis et al., 2023, Mazorra et al., 25 Feb 2026). Across these settings, the common object is not merely blockspace in the abstract, but the ability to convert a transient arbitrage, liquidation, oracle, or post-trade price impact into executable ordering priority.

1. Conceptual scope and relation to MEV microstructure

Backrunning in the narrow MEV sense refers to observing a profitable transaction and inserting one’s own transaction immediately after it to capture induced price movement. Auction-based backrunning generalizes this by focusing on the mechanism through which the post-trade slot, inclusion right, or broader sequencing privilege is allocated. The literature synthesized here distinguishes three analytically adjacent regimes: a block-level auction in which builder bids embed private searcher bundles and orderflow; a fee-auction layer in which strategic commitments alter inclusion pricing; and a timing game in which the right to backrun is effectively purchased through spam and latency-sensitive action placement rather than through an explicit bid scalar (Wu et al., 2023, Landis et al., 2023, Mazorra et al., 25 Feb 2026).

Within post-Merge Ethereum, the most direct block-level object is the MEV-Boost auction. In each 12-second slot, a proposer outsources block construction to specialized builders, builders submit blocks and bids through relays, the relay validates bids and blocks, tracks the highest bid, and returns the header of the highest-paying block when the proposer calls getHeader (Wu et al., 2023). That auction does not sell a single backrun opportunity. Instead, it sells the entire block, whose value is partly composed of private searcher bundles corresponding to “arbitrages, sandwich attacks, liquidations, and cross-domain MEV.” This suggests that auction-based backrunning often appears as a component of a higher-level aggregation market rather than as an isolated transaction-level primitive.

A second layer concerns the transaction fee mechanism itself. When users or searchers can deploy smart contracts that credibly commit them to contingent bidding strategies, the fee auction governing inclusion can be manipulated into a low-revenue equilibrium in which a leader receives inclusion essentially for free and other participants enter a lottery for residual blockspace (Landis et al., 2023). This is not literal DEX-state backrunning, but it is an auction-layer analogue: the mechanism that determines who can compete for ordering is itself strategically exploitable.

A third layer arises when the protocol does not explicitly auction ordering rights at all. In the timing-game model for probabilistic backrunning, an opportunity appears randomly on a time interval, observation is delayed, and whichever player takes an action the fastest after the opportunity has arisen wins (Mazorra et al., 25 Feb 2026). In that environment, repeated costly submissions play the economic role of bids. The allocation rule is first-come-first-served, but the induced competitive structure resembles an all-pay contest.

2. Block-level auctions and builder valuation of backrunning opportunities

The most formal block-auction treatment models the MEV-Boost auction as a continuous-time game among builders indexed by $N=\{1,\ldots,n\}$, where each builder employs a bidding strategy $\beta_s:X\to\mathbb{R}_+$ and submits bids $b_{i,t}=\beta_s(x_{i,t})$ over the interval $[0,T]$ (Wu et al., 2023). The builder’s information consists of a public signal and a private signal. The public component is

$P(t)=\sum\nolimits_{j=1}^{N(t)}V_j,$

with $N(t)\sim \text{Poisson}(\lambda_p\cdot t)$ and $V_j\sim \text{Log-normal}(\xi_1,\omega_1)$. The private component is

$E_i(t)=\sum\nolimits_{j=1}^{N_i(t)}O_j,$

with $$N_i(t)\sim \text{Poisson}(\lambda_e\cdot t\cdot \pi_i)$$ and $O_j\sim \text{Log-normal}(\xi_2,\omega_2)$. The model then defines

$\beta_s:X\to\mathbb{R}_+$0

and the builder’s valuation as

$\beta_s:X\to\mathbb{R}_+$1

Here $\beta_s:X\to\mathbb{R}_+$2 is builder $\beta_s:X\to\mathbb{R}_+$3’s EOF access probability, namely the probability of accessing each searcher bundle, and $\beta_s:X\to\mathbb{R}_+$4 is a builder-specific profit margin (Wu et al., 2023).

This structure is directly relevant to auction-based backrunning because it decomposes willingness to pay for sequencing rights into public mempool value and private bundle access. A privately sourced backrun bundle naturally enters $\beta_s:X\to\mathbb{R}_+$5; a publicly inferable opportunity can be interpreted as part of $\beta_s:X\to\mathbb{R}_+$6. The resulting environment is neither pure independent private values nor pure common values. It is an asymmetric interdependent-value setting with public and private components, in which orderflow access changes valuations before any explicit auction rule is applied.

The payment rule is first-price in effect. If builder $\beta_s:X\to\mathbb{R}_+$7 wins at time $\beta_s:X\to\mathbb{R}_+$8, the payoff is

$\beta_s:X\to\mathbb{R}_+$9

The auction proceeds over continuous time and is discretized to 10ms steps in simulation. No formal Nash equilibrium or Bayes-Nash equilibrium is derived; the analysis is comparative-static and simulation-based. The paper therefore supplies a model of strategic bidding under latency and orderflow asymmetry rather than a closed-form equilibrium theory of transaction-level backrun auctions (Wu et al., 2023).

Several assumptions delimit the interpretation. MEV opportunities are assumed persistent and always profitable throughout the auction; public transactions’ MEV is universal for all builders; searcher payments are uniform among builders who share a bundle; bid cancellations are effective under honest proposers; builders know the current highest bid; and signals are simultaneous in simulation (Wu et al., 2023). These assumptions fit the builder-auction layer more closely than the most ephemeral transaction-level backrun race.

3. Strategic timing, latency, and information asymmetry

The builder-auction model formalizes four bidding strategies: naive, adaptive, last-minute, and bluff (Wu et al., 2023). The naive strategy bids current valuation whenever profitable,

$b_{i,t}=\beta_s(x_{i,t})$0

The adaptive strategy outbids the standing maximum by a small increment $b_{i,t}=\beta_s(x_{i,t})$1 when feasible,

$b_{i,t}=\beta_s(x_{i,t})$2

The last-minute strategy reveals only at the latest feasible moment,

$b_{i,t}=\beta_s(x_{i,t})$3

where, for expected auction end $b_{i,t}=\beta_s(x_{i,t})$4 seconds and revelation slack $b_{i,t}=\beta_s(x_{i,t})$5,

$b_{i,t}=\beta_s(x_{i,t})$6

The bluff strategy bids an inflated amount before $b_{i,t}=\beta_s(x_{i,t})$7 and reverts to true valuation after $b_{i,t}=\beta_s(x_{i,t})$8, exploiting bid cancellation to induce rivals, especially adaptive rivals, to reveal willingness to pay (Wu et al., 2023).

Latency enters through global relay delay $b_{i,t}=\beta_s(x_{i,t})$9 and builder-specific delay $[0,T]$0. These parameters govern whether a reactionary bid is accepted soon enough to matter. The simulation results show that adaptive players’ win rate decreases by $[0,T]$1 for every 10ms increase in global delay, while a 10ms lower individual delay yields about a $[0,T]$2 win-rate advantage (Wu et al., 2023). In a representative threshold example, when global delay is $[0,T]$3ms and individual delay is $[0,T]$4ms for adaptive players, total latency is $[0,T]$5ms; if last-minute revelation occurs less than 50ms before the end, adaptive players cannot react in time. Auction efficiency, defined as the ratio between the winning bid value and the total signal, decreases by $[0,T]$6 per 10ms increase in global delay (Wu et al., 2023).

These results are central for auction-based backrunning because the mechanism need not allocate rights to the highest-value participant; it can allocate them to the participant who can convert value into an accepted bid inside the residual latency window. Naive bidding remains surprisingly competitive because it is simple and not latency-fragile. Adaptive bidding can improve profit-per-win but loses win rate under latency. Last-minute revelation is highly effective in fixed-timing auctions because it suppresses response. Bluffing mainly manipulates adaptive players but becomes dangerous when auction end times are uncertain (Wu et al., 2023). A plausible implication is that any open, continuously updating auction for backrun rights inherits a timing game over reaction windows in addition to the valuation contest itself.

Latency, however, is not the only or even the dominant asymmetry. EOF access has a larger, nonlinear effect on performance. In the reported experiments, a 10ms latency advantage gives about $[0,T]$7 better win rate and $[0,T]$8 ETH more average profit, whereas increasing EOF access probability leads to exponential increases in performance (Wu et al., 2023). This suggests that private orderflow and privileged searcher-builder relationships may dominate raw network speed in determining auction power.

4. Commitment attacks and manipulation of the auction layer

A distinct line of work studies auctions with $[0,T]$9 identical items and $P(t)=\sum\nolimits_{j=1}^{N(t)}V_j,$0 agents, where agents can deploy smart contracts that credibly commit them to contingent strategies (Landis et al., 2023). In blockchain terms, the items are block slots, the agents are transactions, users, or searchers competing for inclusion, and the strategic novelty is a metagame of commitments. The paper asks whether standard auction mechanisms are Stackelberg resilient, meaning that the addition of commitment moves leaves equilibrium payoffs weakly strategically equivalent to the original game. Its main conclusion is negative for first-price auctions, second-price auctions, and Ethereum’s EIP-1559 (Landis et al., 2023).

In the first-price benchmark with complete information, the top $P(t)=\sum\nolimits_{j=1}^{N(t)}V_j,$1 bidders win and pay their own bids, with a standard equilibrium

$P(t)=\sum\nolimits_{j=1}^{N(t)}V_j,$2

Under EIP-1559, each agent submits a transaction with value $P(t)=\sum\nolimits_{j=1}^{N(t)}V_j,$3 and deposit $P(t)=\sum\nolimits_{j=1}^{N(t)}V_j,$4, the miner chooses $P(t)=\sum\nolimits_{j=1}^{N(t)}V_j,$5 with $P(t)=\sum\nolimits_{j=1}^{N(t)}V_j,$6 maximizing $P(t)=\sum\nolimits_{j=1}^{N(t)}V_j,$7, included agent $P(t)=\sum\nolimits_{j=1}^{N(t)}V_j,$8 receives utility $P(t)=\sum\nolimits_{j=1}^{N(t)}V_j,$9, and miner revenue is $N(t)\sim \text{Poisson}(\lambda_p\cdot t)$0 (Landis et al., 2023).

The core attack is the contract family $N(t)\sim \text{Poisson}(\lambda_p\cdot t)$1. If all agents adopt the prescribed contracts, the leader bids $N(t)\sim \text{Poisson}(\lambda_p\cdot t)$2, non-coalition cooperators bid $N(t)\sim \text{Poisson}(\lambda_p\cdot t)$3, and the remaining $N(t)\sim \text{Poisson}(\lambda_p\cdot t)$4 slots are allocated uniformly among the $N(t)\sim \text{Poisson}(\lambda_p\cdot t)$5 non-coalition agents. The leader is therefore guaranteed inclusion at cost $N(t)\sim \text{Poisson}(\lambda_p\cdot t)$6, and each non-coalition participant receives lottery winning probability

$N(t)\sim \text{Poisson}(\lambda_p\cdot t)$7

and expected utility

$N(t)\sim \text{Poisson}(\lambda_p\cdot t)$8

If anyone deviates, the contracts revert to the competitive auction profile around cutoff $N(t)\sim \text{Poisson}(\lambda_p\cdot t)$9, making defection unattractive when valuations are sufficiently concentrated (Landis et al., 2023). The viability condition is

$V_j\sim \text{Log-normal}(\xi_1,\omega_1)$0

Under this condition, EIP-1559 is not Stackelberg attack resilient; the same asymptotic condition is stated for second-price auctions, and the first-price case is obtained by setting $V_j\sim \text{Log-normal}(\xi_1,\omega_1)$1 (Landis et al., 2023).

For random valuations, the paper reports high-probability results. If valuations are i.i.d. uniform and $V_j\sim \text{Log-normal}(\xi_1,\omega_1)$2, first-price auctions are not Stackelberg resilient, except with negligible probability in $V_j\sim \text{Log-normal}(\xi_1,\omega_1)$3, whenever $V_j\sim \text{Log-normal}(\xi_1,\omega_1)$4; the derived numerical threshold is $V_j\sim \text{Log-normal}(\xi_1,\omega_1)$5. For i.i.d. Pareto($V_j\sim \text{Log-normal}(\xi_1,\omega_1)$6) valuations with $V_j\sim \text{Log-normal}(\xi_1,\omega_1)$7, the attack works except with probability negligible in $V_j\sim \text{Log-normal}(\xi_1,\omega_1)$8 whenever $V_j\sim \text{Log-normal}(\xi_1,\omega_1)$9, with $E_i(t)=\sum\nolimits_{j=1}^{N_i(t)}O_j,$0 (Landis et al., 2023).

The economic effect is a revenue collapse for the auctioneer. In the $E_i(t)=\sum\nolimits_{j=1}^{N_i(t)}O_j,$1 example, ordinary first-price revenue is $E_i(t)=\sum\nolimits_{j=1}^{N_i(t)}O_j,$2, attacked revenue is $E_i(t)=\sum\nolimits_{j=1}^{N_i(t)}O_j,$3, and the auctioneer loses almost all revenue. Participant welfare can rise even as seller revenue falls; for uniformly distributed valuations the reported “price of defiance” satisfies

$E_i(t)=\sum\nolimits_{j=1}^{N_i(t)}O_j,$4

with high probability (Landis et al., 2023). For auction-based backrunning, this matters because it demonstrates that the auction layer allocating inclusion can be collusively or commitment-theoretically manipulated even before one reaches the microstructure of bundle ordering.

5. Implicit auctions: probabilistic backrunning as a costly timing contest

Probabilistic backrunning is modeled as a timing game $E_i(t)=\sum\nolimits_{j=1}^{N_i(t)}O_j,$5 with $E_i(t)=\sum\nolimits_{j=1}^{N_i(t)}O_j,$6 players, opportunity value $E_i(t)=\sum\nolimits_{j=1}^{N_i(t)}O_j,$7, marginal cost $E_i(t)=\sum\nolimits_{j=1}^{N_i(t)}O_j,$8 per action, and random opportunity time $E_i(t)=\sum\nolimits_{j=1}^{N_i(t)}O_j,$9 distributed according to an absolutely continuous, strictly increasing CDF $$N_i(t)\sim \text{Poisson}(\lambda_e\cdot t\cdot \pi_i)$$0 with density $$N_i(t)\sim \text{Poisson}(\lambda_e\cdot t\cdot \pi_i)$$1 (Mazorra et al., 25 Feb 2026). A pure strategy of player $$N_i(t)\sim \text{Poisson}(\lambda_e\cdot t\cdot \pi_i)$$2 is a finite subset $$N_i(t)\sim \text{Poisson}(\lambda_e\cdot t\cdot \pi_i)$$3, interpreted as the set of times at which the player sends transactions. Whoever has the first action after $$N_i(t)\sim \text{Poisson}(\lambda_e\cdot t\cdot \pi_i)$$4 wins. Formally, using directed distance

$$N_i(t)\sim \text{Poisson}(\lambda_e\cdot t\cdot \pi_i)$$5

the player wins when its first post-$$N_i(t)\sim \text{Poisson}(\lambda_e\cdot t\cdot \pi_i)$$6 action minimizes $$N_i(t)\sim \text{Poisson}(\lambda_e\cdot t\cdot \pi_i)$$7. The payoff is

$$N_i(t)\sim \text{Poisson}(\lambda_e\cdot t\cdot \pi_i)$$8

The model is motivated by cases in which an AMM trade, oracle update, or similar state change creates a backrunning opportunity, but the event is observed only with delay (Mazorra et al., 25 Feb 2026).

This is not an explicit auction, but it has an auction-like interpretation. The paper states that the competitive structure resembles an all-pay contest: all players incur costs through their actions, but only the earliest successful action captures the prize (Mazorra et al., 25 Feb 2026). The timing game admits no pure Nash equilibrium. For $$N_i(t)\sim \text{Poisson}(\lambda_e\cdot t\cdot \pi_i)$$9 and $O_j\sim \text{Log-normal}(\xi_2,\omega_2)$0, it has an almost surely unique symmetric Nash equilibrium, and each player’s equilibrium payoff is zero. Equilibrium strategies have a recursive structure: there exist i.i.d. random variables $O_j\sim \text{Log-normal}(\xi_2,\omega_2)$1 and a map $O_j\sim \text{Log-normal}(\xi_2,\omega_2)$2 such that

$O_j\sim \text{Log-normal}(\xi_2,\omega_2)$3

The support begins at $O_j\sim \text{Log-normal}(\xi_2,\omega_2)$4, any distinct action times in a pure best reply satisfy spacing at least $O_j\sim \text{Log-normal}(\xi_2,\omega_2)$5, and the support of the intensity measure is the full interval $O_j\sim \text{Log-normal}(\xi_2,\omega_2)$6 (Mazorra et al., 25 Feb 2026).

A key equilibrium condition is that every singleton deviation breaks even:

$O_j\sim \text{Log-normal}(\xi_2,\omega_2)$7

Equivalently,

$O_j\sim \text{Log-normal}(\xi_2,\omega_2)$8

For two players in the high-cost one-shot regime $O_j\sim \text{Log-normal}(\xi_2,\omega_2)$9, each player sends at most one transaction in equilibrium, with first-layer distribution

$\beta_s:X\to\mathbb{R}_+$00

The aggregate welfare statement is stark. Let $\beta_s:X\to\mathbb{R}_+$01 be the event that someone captures the opportunity. Then

$\beta_s:X\to\mathbb{R}_+$02

where

$\beta_s:X\to\mathbb{R}_+$03

Thus expected total fees or spam cost exactly equals the probability that the opportunity is captured in equilibrium (Mazorra et al., 25 Feb 2026). The paper also reports

$\beta_s:X\to\mathbb{R}_+$04

For $\beta_s:X\to\mathbb{R}_+$05, numerical examples give spam around $\beta_s:X\to\mathbb{R}_+$06 at $\beta_s:X\to\mathbb{R}_+$07 and slowly falling toward $\beta_s:X\to\mathbb{R}_+$08 by $\beta_s:X\to\mathbb{R}_+$09, with lower bound $\beta_s:X\to\mathbb{R}_+$10 (Mazorra et al., 25 Feb 2026). This identifies a non-auction benchmark for backrunning: if ordering is FCFS under delayed information, the protocol implements an implicit all-pay auction in which the bid is the total cost of spam.

6. Design implications, limitations, and recurring controversies

Three recurring design lessons emerge from the literature. First, latency materially shapes outcomes. In dynamic open auctions with visible standing bids, reactive minimal-increment strategies are fragile, late revelation can dominate, and seller revenue falls when late value cannot be incorporated before close (Wu et al., 2023). Second, private orderflow can matter more than modest latency advantages. Builder-specific EOF access changes valuations themselves, not merely execution speed, and the reported performance effect is nonlinear (Wu et al., 2023). Third, when the protocol relies on FCFS rather than explicit pricing, the right to backrun is not removed from competition but reallocated into spam and rent dissipation with zero expected player profit (Mazorra et al., 25 Feb 2026).

The literature also identifies manipulation channels beyond latency. If participants can commit through smart contracts to contingent bidding strategies, first-price auctions, second-price auctions, and EIP-1559 are not Stackelberg resilient; under the stated concentration conditions, competition can collapse into a lottery-like low-fee regime that benefits participants while destroying proposer or miner revenue (Landis et al., 2023). This is especially relevant for auction-based backrunning because the inclusion mechanism itself can become the object of extraction.

Several common misconceptions require qualification. One is that existing papers provide a direct theory of transaction-level backrun auctions. They do not. The MEV-Boost model studies builder-level block auctions, not auctions for a single backrun slot; the Stackelberg model studies fee-market manipulation rather than DEX post-trade execution; and the timing-game model studies probabilistic backrunning without an explicit auctioneer (Wu et al., 2023, Landis et al., 2023, Mazorra et al., 25 Feb 2026). Another misconception is that “faster always wins.” The builder-auction results show that access to private orderflow can dominate modest latency improvements, while the timing-game results show that low-cost competition can dissipate value into spam regardless of strategic sophistication (Wu et al., 2023, Mazorra et al., 25 Feb 2026). A further misconception is that open auction formats necessarily improve efficiency. The builder-auction study finds that bid cancellation enables bluffing, fixed visible cutoffs enable sniping, and uncertain close times can punish last-second tactics but do not constitute a clean solution (Wu et al., 2023).

The principal limitations are likewise structural. The builder-auction model assumes persistent and always profitable opportunities, simultaneous signal updates, and no searcher-level strategic layer. The Stackelberg model assumes strong commitment expressiveness and near-complete information about valuations. The timing-game model assumes symmetric players, common-value opportunity size $\beta_s:X\to\mathbb{R}_+$11, FCFS-style ordering, and no explicit builder or validator strategy (Wu et al., 2023, Landis et al., 2023, Mazorra et al., 25 Feb 2026). These are not minor technicalities; they determine which dimensions of real-world backrunning are being captured.

Taken together, the literature suggests that auction-based backrunning is not a single mechanism but a family of allocation problems over sequencing rights. In some systems the relevant auction is explicit and block-level, with backrunning value embedded in builder bids. In others it is a fee-market mechanism vulnerable to contingent commitment. In still others it is an implicit all-pay contest implemented through spam. The unifying theme is that profitable post-trade positioning is priced through strategic competition over ordering, and the decisive variables are not only the raw opportunity value but also timing, observability, cancellation rules, orderflow exclusivity, and the protocol’s method of closing competition (Wu et al., 2023, Landis et al., 2023, Mazorra et al., 25 Feb 2026).