Research
How To Cook The Fragmented Rug Pull?
Existing rug pull detectors look for a simple signature: the deployer keeps the liquidity pool (LP) tokens and then performs one or a few large sells or withdrawals, usually within a day, that…

- arXiv
- 2511.15463
- Published
- 2025-11-19
- Authors
- Minh Trung Tran, Nasrin Sohrabi, Zahir Tari, Qin Wang
AI summary
Overview
Research area: Blockchain security — on-chain fraud detection, specifically rug pull scams on Ethereum decentralized exchanges (DEXs).
Technical level: Intermediate. The paper combines formal predicate logic with large-scale on-chain measurement; the core ideas are graspable with basic DeFi knowledge (liquidity pools, LP tokens, automated market makers), while the formal definitions and dataset construction require more familiarity.
Scope (one sentence): The paper formalizes "fragmented rug pulls" — exits split into many low-impact trades routed through many non-owner wallets — and measures their prevalence across six Ethereum DEXs from November 2018 to March 2025.
What This Paper Is About
Existing rug pull detectors look for a simple signature: the deployer keeps the liquidity pool (LP) tokens and then performs one or a few large sells or withdrawals, usually within a day, that collapse the pool. The authors show that real attackers increasingly break the same total extraction into many small trades and route them through multiple wallets that are not the owner, so no single transaction or address trips the old rules. The goal is to formally define this evasive attack family, show why the classic heuristics are structurally brittle, and quantify how widespread the pattern actually is on Ethereum.
Key Contributions
-
Predicate decomposition of rug pull heuristics. The authors break the canonical sell-based detector into three atomic predicates — (A) LP control retention, (B) a single-transaction impact above a threshold θ, and (C) the seller being in the owner set — and analyze which ones an attacker can change without reducing payoff.
-
A formal definition of the fragmented rug pull (FRP). The paper defines FRP as an exit strategy parameterized by the number of selling wallets (N), the per-wallet order count (K_a), and per-order volume (v), subject to every order staying at or below θ and no heuristic-triggering owner footprint.
-
A large-scale on-chain measurement. Applying the definition to 303,614 deployed LPs (retained from 384,029 pools across six DEXs), the authors label 105,434 pools as FRP and characterize who performs the drains and how.
-
A released labeled dataset and analysis artifacts. The authors state they release their labeled pool dataset, predicate evaluations, analysis scripts, and notebooks at an anonymous.4open.science repository, plus practical early-warning indicators.
Main Findings
-
The classic heuristic is brittle by construction. Only predicate (A) — retaining LP control — is argued to be necessary for profitability; the paper states that violating it yields an on-chain return strictly smaller than the canonical return. Predicates (B) and (C) can each be modified independently while preserving an equivalent or "satisficing" profit.
-
FRPs are widespread. Of 303,614 short-lived pools (lifetime ≤ 100 days), 105,434 — reported as 34.7% — are labeled as fragmented rug pulls, involving 34,192,767 pool-related transactions, 401,838 inflated-seller wallets, and 1,501,408 unique interacting addresses.
-
Drain activity has moved off the owner wallet. The abstract reports that owner-wallet participation in inflated selling among FRP-flagged pools has declined substantially, at approximately 33.1% of cases, indicating the liquidity drain is no longer held on the owner wallet. The introduction separately reports owner participation dropping from 65% to 24%.
-
Multi-wallet schemes are now the norm. Traditional single-wallet exits fell from 57% in 2019 to 28.3% in 2024, and over 70% of FRPs now involve coordinated actions from a few sellers rather than one.
-
Serial scamming is measurable. The authors detected 127,252 wallets acting as serial scammers by repeatedly engaging in inflated selling across multiple FRP LPs.
-
Adoption exploded after 2020. FRP presence was negligible before 2020 (fewer than one hundred cases observed) but reached nearly 40k detected pools in 2023.
-
Exit styles differ measurably. Single-wallet scams finish within a couple of days with only a handful of sells, whereas multi-wallet schemes stretch over a week with dozens of small trades that mimic normal market activity.
-
Motivating real cases evade automated detectors. LIBRA reached over $5 billion in token value within a few hours of launch but exited through four wallets created within one hour of deployment, each making only two to five medium-sized sells, every one below the θ ≈ 0.9 threshold. Safereum (about a dozen pre-funded wallets, profits consolidated and bridged out) and ZhongHua (a proxy contract making timed sells through delegated EOAs) show related variants; the paper states existing automated detectors missed all of them.
Methodology in Plain English
The authors start from how existing detectors work and rewrite those rules as three small, checkable conditions. They then reason about what happens if an attacker keeps the first condition (holding the LP tokens) but changes the other two: sell in smaller pieces so no trade looks big, and sell from wallets that are not the deployer's. They show mathematically that neither change removes the ability to profit, because splitting a trade can actually reduce slippage cost faster than it adds gas fees.
To test whether this happens in reality, they pull all Ethereum transactions from block 6,627,000 (Nov. 2, 2018) to block 21,379,910 (Mar. 5, 2025) using a Geth client node, cover six DEXs (Uniswap, SushiSwap, Balancer, Curve, PancakeSwap, BancorSwap), and match deployments using the Chainalysis API. They categorize every interaction as deploy, buy, sell, deposit, or withdraw, drop pools whose observable lifetime exceeds 100 days, and then apply their FRP definition with θ set to 0.9 — also sweeping θ from 0.7 to 0.95 to test sensitivity. Finally, they analyze the flagged pools in two directions: actor-centric (who sells, whether the owner participates, whether wallets recur across pools) and action-centric (how many sells, how fragmented, what behavioral category).
Why This Matters
For research, the paper reframes rug pull detection as a problem of reasoning over aggregate behavior rather than isolated transactions, and it supplies a labeled dataset and formal vocabulary that later work can build on or contest.
Real-world applications:
- Detector design for DEX analytics platforms and security firms: shift from per-transaction thresholds to wallet-cluster and cumulative-drain signals.
- Investor and trader warning tools: flag early indicators such as multiple distributor wallets created shortly after deployment, synchronized low-impact sells, and rapid post-exit fund consolidation.
- Exchange, bridge, and custody compliance screening: identify wallets with serial inflated-selling histories across many pools (the 127,252 recurring wallets reported here).
- LP-lock and launchpad due diligence: treat retained LP control combined with distributed selling as a risk pattern rather than relying solely on burn/lock status.
Industry relevance is substantial: the paper notes Ethereum holds roughly 70% of total value locked across blockchains, and that rug pulls cause annual losses exceeding $2B USD.
Future Directions
-
Detection methods that operate on aggregated behavior. The authors explicitly call for detectors that reason over cumulative multi-wallet activity instead of isolated events, which their measurement motivates but does not itself build as a deployed system.
-
Ownership inference under obfuscation. The paper notes that identity delegation forces defenders to infer ownership indirectly through behavioral clustering or timing analysis; how reliably that can be done is left open.
-
Semantic ownership hiding. The authors explicitly exclude cases where control is transferred to a DAO-like multisig or timelock contract while trading rights are retained elsewhere, leaving these ambiguous ownership structures unaddressed.
-
Cross-chain variants. Cross-wallet and cross-chain laundering is described as the most sophisticated delegation tier, but the measurement is confined to Ethereum DEXs, so how FRP patterns appear on other chains is not reported.
Target Audience
Blockchain security researchers and detector engineers, DeFi analysts and on-chain forensics teams, and graduate students working on measurement-driven fraud studies. It is also useful for technically literate investors and compliance staff who want to understand why owner-sell and large-trade alerts can miss real exits, though readers without familiarity with liquidity pools and AMM mechanics will need background reading first.
Authors’ abstract
Existing rug pull detectors assume a simple workflow: the deployer keeps liquidity pool (LP) tokens and performs one or a few large sells (within a day) that collapse the pool and cash out. In practice, however, many real-world exits violate these assumptions by splitting the attack across both time and actor dimensions: attackers break total extraction into many low-impact trades and route proceeds through multiple non-owner addresses, producing low-visibility drains. We…
The full abstract on arXiv