💻 computer science

To Wait or To Probe: Arbitrage Competition on High-Throughput Blockchains

This paper analyzes how high-throughput blockchain configurations influence the competition between targeted and probabilistic MEV search strategies, demonstrating that while probabilistic search constitutes a minority of arbitrage activity, it generates the vast majority of spam and gas consumption, whereas protocol adjustments like fee floors and flashblocks effectively shift revenue toward successful trades and reduce network congestion.

Original authors: Fei Wu, Burak Öz

Published 2026-06-02

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Fei Wu, Burak Öz

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a high-speed digital marketplace where people are racing to find and buy rare items (like a first-edition book) the moment they appear. In this paper, the authors study how two different types of "hunters" (bots) compete to find these opportunities on fast blockchains like Base, and how the rules of the game change who wins.

Here is the breakdown of the paper using simple analogies:

The Two Types of Hunters

The paper identifies two main strategies bots use to find profit:

The "Planner" (Targeted Search):
- How they work: This bot sits back and waits. It uses powerful computers outside the marketplace to figure out exactly where the rare book is, what it costs, and how to buy it. Once it has a perfect plan, it walks into the store and makes one single, confident purchase.
- The result: They are very efficient. They rarely make mistakes, but they need to be sure of the information before they act.
The "Sweeper" (Probabilistic Search):
- How they work: This bot is impatient. It doesn't wait for a perfect plan. Instead, it runs into the store and starts frantically checking every shelf, every aisle, and every bin, thousands of times a second. It asks, "Is the book here? Is it there?" If it finds a match, it buys it. If not, it keeps checking.
- The result: They find opportunities faster sometimes, but they create a massive amount of "noise." Most of their checks are useless, clogging up the store's aisles and making it hard for other customers to move. In the paper, these useless checks are called spam.

The Problem: The Store is Getting Crowded

The authors noticed that while "Sweepers" only account for about 23% of the actual successful deals, they are responsible for 95% of the noise (spam) and use up 20% of the store's total space. It's like a library where one person is checking out a book, but 100 other people are running around shouting "Is the book here?" over and over, blocking the doors.

The Experiments: Changing the Rules

The researchers watched what happened when the "store managers" (the blockchain protocol) changed the rules in three specific ways:

1. The "Flashblocks" Experiment (Faster, Smaller Windows)

The Change: The store started updating its inventory list every 200 milliseconds instead of every 2 seconds. This created tiny, fast-moving windows of opportunity.
The Effect: The "Sweepers" struggled. Because the windows were so small, the heavy, slow bots that checked too many shelves got pushed out of the line. Only the leaner, faster bots survived.
The Outcome: The number of "Sweeper" bots dropped dramatically. However, the ones that did survive started checking shelves even more frantically to compensate. So, while there were fewer Sweepers, the total noise didn't drop as much as hoped because the survivors were working harder.

2. The "New Token" Surge (A Temporary Gold Rush)

The Change: Two new, popular items (tokens named AVNT and MIRROR) were suddenly introduced. No one knew the rules for them yet.
The Effect: This was a temporary boom for the "Sweepers." Since no one had a pre-made plan for these new items, the bots that could run around and check everything in real-time had an advantage. A huge wave of new bots rushed in to try their luck.
The Outcome: This created a massive spike in spam. But as soon as the new items became "normal" and people figured out the rules, the "Sweepers" left, and the "Planners" took over again.

3. The "Fee Hike" (Making it Costly to Check)

The Change: The store raised the minimum fee required just to walk through the door and check a shelf.
The Effect: This hurt the "Sweepers" the most. Since they check thousands of shelves and fail most of the time, the cost of walking through the door added up quickly. If a bot checks 1,000 shelves and only finds one deal, but the fee for checking is too high, they lose money.
The Outcome: The bots that couldn't afford to keep checking (the low-value ones) left the store. The "Sweepers" that stayed were the ones with the most valuable deals, but their numbers shrank significantly. The "Planners," who only make one check per deal, were less affected.

The Big Takeaway

The paper concludes that protocol design matters.

If you make the ordering of transactions very fast and granular (like Flashblocks), you naturally push out the heavy, spammy bots, but the survivors might just work harder.
If you raise the fees, you force the bots to be more efficient; they can't afford to waste time checking things that aren't valuable.

Ultimately, the authors show that you can't just "ban" spam. You have to understand how the bots are hunting. By changing the rules of the game (speed, fees, and information), the blockchain can encourage the "Planners" and discourage the "Sweepers," making the marketplace cleaner and more efficient for everyone.

Technical Summary: To Wait or To Probe: Arbitrage Competition on High-Throughput Blockchains

1. Problem Statement

On high-throughput blockchains, Maximal Extractable Value (MEV) search has bifurcated into two distinct economic architectures: targeted search and probabilistic search.

Targeted Search: Bots identify opportunities off-chain, simulate execution, and submit route-committed transactions. This approach typically involves fewer attempts but higher fees per attempt.
Probabilistic Search: Bots submit repeated transactions that read live state on-chain to discover opportunities during execution. This approach involves high attempt intensity, where most transactions fail (probes or reverts) without producing useful state changes, constituting "spam."

The core problem is how protocol design parameters—specifically ordering granularity (e.g., sub-blocks vs. full blocks), fee floors, and opportunity access—shape the competition between these architectures. While existing literature models probabilistic search in isolation, this paper formalizes the distinction between the two architectures and investigates how protocol changes shift the equilibrium composition of bots, blockspace consumption, and protocol revenue.

2. Methodology

Theoretical Framework

The authors develop a reduced-form equilibrium model of architecture competition. The model assumes free entry until expected profits are driven to zero. Key variables include:

Ordering Regimes: Baseline (full block) vs. Finer (sub-block/slice) ordering.
Competitiveness ( $\eta_r$ ): The probability that an on-chain discovery transaction remains competitive in the same profitable slice as an off-chain transaction. The model posits that finer ordering reduces $\eta_r$ for on-chain discovery due to higher blockspace consumption per attempt.
Opportunity Access ( $\lambda$ ): The rate at which bots can access opportunities. On-chain discovery generally has higher access ( $\lambda_C \ge \lambda_O$ ) as it can scan multiple routes dynamically.
Costs ( $K$ ): Total costs including gas and fees, which vary by architecture and fee floors.

The model derives comparative statics predicting that:

Finer ordering reduces the viability of on-chain discovery, particularly for bots with high "scan footprints" (heavy state reads).
Higher fee floors disproportionately hurt architectures with high fee exposure (high gas per attempt and high attempt intensity).
Opportunity shocks (e.g., new token launches) can temporarily revive on-chain discovery by creating routes not yet indexed off-chain.

Empirical Strategy

The study utilizes Base as a case study, covering the period from June 2025 to February 2026. This period includes major configuration changes: the introduction of Flashblocks (200ms sub-blocks) and a multi-step escalation of the minimum base fee.

Data Construction:

Dataset: 21.37 million successful cyclic arbitrages and 985 million spam transactions (probes and reverts) from 4,365 unique bot addresses.
Classification: A trace-level classifier distinguishes bot architectures based on successful arbitrage transactions. It analyzes debug_traceTransaction outputs to categorize bots into:
- Off-chain Discovery: Route committed off-chain; no pre-swap state reads or reads confined to the execution route.
- On-chain Evaluation: Reads state on-chain but confined to the eventual execution route.
- On-chain Discovery (Probabilistic): Scans a superset of venues before selecting a route (broad on-chain discovery).
Aggregation: Labels are aggregated at the bot-week level to analyze population shifts, spam intensity, and efficiency metrics.

Analysis Episodes

The empirical analysis tests the model's predictions across three specific episodes:

Flashblocks Activation (July 2025): Tests the effect of finer ordering.
September Surge (Sept 2025): Tests the effect of opportunity-access shocks (new token launches AVNT/MIRROR).
Minimum Base Fee Escalation (Dec 2025 – Feb 2026): Tests the effect of rising fee floors.

3. Key Contributions

Formalization of Search Architectures: The paper formally distinguishes between targeted and probabilistic search, modeling how ordering granularity and fees shape their relative viability.
Trace-Level Classifier: Development of a novel classifier that identifies search architectures based on observable route discovery in transaction traces, validated against execution behavior and spam profiles.
Empirical Validation of Comparative Statics:
- Flashblocks: Confirmed that finer ordering selects against high-scan-footprint on-chain discovery bots, shifting the market toward leaner, off-chain discovery bots.
- Opportunity Shocks: Demonstrated that new token launches temporarily revive probabilistic search, driven by short-lived entrants.
- Fee Floors: Showed that higher fee floors select against bots whose opportunity flow cannot sustain repeated attempts, reducing the viability of high-exposure architectures.
Quantification of Chain-Level Outcomes: Measured the impact of these shifts on priority fee revenue composition and blockspace efficiency.

4. Key Results

Bot Architecture Characteristics

Probabilistic Search (On-chain Discovery): Accounts for only 23% of successful arbitrage activity but generates 95% of spam transactions and consumes 20% of Base's total gas. These bots have low success rates (4.58%) and high costs per success.
Targeted Search (Off-chain Discovery): Higher success rates (42.51%), lower spam generation, and higher priority fee payments per transaction.

Episode-Specific Findings

Flashblocks (Finer Ordering):
- Selection Effect: The population of on-chain discovery bots dropped from ~63% to ~25% of active bots. High-scan-footprint bots (short calldata, high pre-reads) were selected out.
- Survivors: Remaining on-chain bots simplified their logic (longer calldata, fewer reads).
- Spam Paradox: Despite the reduction in bot population, aggregate spam did not fall proportionally because surviving bots increased their attempt intensity (probing more frequently).
- Efficiency: Off-chain discovery success rates improved (35.8% $\to$ 43.6%) due to reduced stale-state risk.
September Surge (Opportunity Shock):
- A temporary spike in on-chain discovery bots (from ~30 to ~220) occurred around the launch of AVNT and MIRROR tokens.
- Transient Entrants: 92% of these new bots were short-lived "triers" that exited once the specific token routes were routinized.
- Spam Shock: This wave generated 147M spam transactions, mostly from transient bots.
Fee Floor Escalation (Cost Channel):
- As the minimum base fee rose from 200K to 5M WEI, the share of on-chain discovery bots fell from 22% to 8.4%.
- Selection Mechanism: The fee floor selected against bots with low value-per-attempt. Bots survived only if their successful opportunities were valuable enough to offset the cumulative cost of repeated failed attempts.
- Footprint Stability: Unlike Flashblocks, fee hikes did not select for less scan-intensive logic; rather, they pruned low-yield attempts.

Protocol Revenue and Blockspace

Revenue Composition: Post-Flashblocks, the share of priority fees derived from spam dropped from >60% to ~40-45%. Revenue shifted toward successful arbitrages and higher-bidding probabilistic attempts.
Blockspace Efficiency: While spam remains a significant burden, its share of total gas consumption declined after Flashblocks and further during the fee ramp (from ~20% to ~15%), aside from temporary spikes during opportunity shocks.

5. Significance and Claims

The paper claims that protocol design does not merely affect the volume of MEV extraction but fundamentally alters the mechanism of search.

Equilibrium Behavior: Failed attempts (spam) are not merely inefficiencies but equilibrium outcomes of search architectures under specific ordering and fee constraints.
Policy Trade-offs:
- Finer Ordering (e.g., Flashblocks) effectively selects against broad scanning but may induce higher attempt intensity among survivors, limiting total spam reduction.
- Fee Floors act as a filter for value-cost margins, disproportionately impacting probabilistic searchers with low-yield opportunities.
Externalities: The primary externality of probabilistic search is blockspace congestion. Protocol changes can mitigate this by shifting the bot composition toward targeted search, though temporary opportunity shocks can revive spam.

The authors conclude that understanding these architectural distinctions is crucial for designing protocols that reduce spam externalities without suppressing useful economic activity like arbitrage. They suggest that future work should explore how these dynamics apply to other MEV forms (liquidations, cross-domain arbitrage) and different blockchain environments.