# Meme Coin Factories: Uncovering Large-Scale Manipulations on pump.fun

Crypto.world

> Creating a cryptocurrency on a launchpad like pump.fun now takes a few clicks and almost no money, which has produced millions of “meme coins” that carry no real technological function.

**arXiv**: 2609.10246

**Published**: 2026-09-09

**Authors**: Nicolas Szwajcok, Taro Tsuchiya, Enze Liu, Kyle Soska, Mathias Payer, Nicolas Christin

## AI summary

## Overview

**Research area:** Blockchain security and measurement — market manipulation on Solana-based meme coin launchpads (cs.CR, with a social-aspects-of-security framing).

**Technical level:** Intermediate. The paper is a large-scale measurement study; the concepts (wash trading, pump-and-dump, address clustering) are explainable without deep cryptography, but the analysis relies on transaction-level blockchain data and statistical modeling.

**Scope (1 sentence):** The paper measures, across all 15,245,966 coins launched on pump.fun between January 14, 2024 and January 14, 2026, five classes of market-manipulation strategies plus the emergence of “Market-Manipulation-as-a-service” (MMaaS) tooling.

## What This Paper Is About

Creating a cryptocurrency on a launchpad like pump.fun now takes a few clicks and almost no money, which has produced millions of “meme coins” that carry no real technological function. The same ease lets strategic actors fake demand signals — trading volume, apparent ownership, coin names, and social media buzz — so that inexperienced traders buy in and the manipulators can sell at inflated prices. The paper’s goal is to detect these manipulations at scale, characterize the infrastructure behind them, and propose mitigations for traders, the platform, wallets and chain scanners, software development platforms, and regulators.

## Key Contributions

1. **A much larger, uniformly sampled measurement than prior work.** The authors track metadata for all 15,245,966 pump.fun coins and collect full transaction histories for a random 1% coin sample (152,171 coins, 49,970,921 transactions) and a 5-day sample (149,028 coins, 37,350,061 transactions) — over 87 million transactions and full transaction data for over 300,000 coins. They contrast this with the closest prior study (Ding et al.), which analyzed about 6,000 coins (0.04% of all coins).
2. **A five-class taxonomy of manipulation strategies** — wash trading, creator address obfuscation, coordinated sell, copycat coins, and social media manipulation — each with detection heuristics (WT1/WT2 for wash trading, DP1/DP2 for coordinated sell) and reported prevalence.
3. **Novel analyses the authors describe as firsts:** creator obfuscation via funding-graph clustering, social media *post* manipulation (posts that spawn new coins), and MMaaS tooling.
4. **Mitigation recommendations** targeting five stakeholder groups: traders, pump.fun, wallets/chain scanners, software development platforms, and regulators (the abstract and introduction state these recommendations; the detailed mitigations section is not included in the provided excerpt).

## Main Findings

- **Wash trading is widespread and mostly non-UI.** A conservative lower bound (WT1, atomic same-transaction buy-and-sell of the same amount against the same bonding curve) yields 4 million wash-trading transactions across the two samples (2,221,734 in the 1% coin sample; 1.8 million in the 5-day sample), equal to 17% of all trading transactions. Because pump.fun’s website does not allow buying and selling in a single transaction, WT1 captures only activity executed outside the interface, indicating technical sophistication.
- **Wash trading concentrates in high-activity coins.** The wash-trading ratio rises from 0.41% for coins with 1–10 transactions (58,084 coins) to 21.65% for coins with 1,001–10,000 transactions (2,295 coins) and 50.31% for coins with more than 10,000 transactions (27 coins).
- **Wash trading is small in volume but large in count.** WT1 accounts for 66,636 SOL, or 1.32% of total trading volume (median 1.00% per coin), suggesting manipulators prioritize transaction count over volume because pump.fun charges fees on trading volume.
- **Wash trading correlates with graduation.** The graduation rate is 2.0% for wash-traded coins versus 0.90% for non-wash-traded coins, rising to 9.64% for coins with 500+ wash-trading transactions (985 coins). A logistic regression finds that doubling the number of wash-trading transactions raises graduation odds by about 19% (p = 3 × 10⁻⁵⁹); the introduction frames the potential effect as up to a 10× increase in success probability. Only 1.02% of coins in the sample graduate at all.
- **Broader wash-trading definitions inflate coin counts fast.** WT2 with a 1-second window flags 18,730 coins (12.38%); a 5-second window flags 2,356,189 instances (44.05% of coins); a 1-day window flags 88.5% of coins — which the authors read as evidence of false positives rather than more manipulation.
- **Creator identities are heavily obfuscated.** From 5,826,346 unique creator addresses, 1-hop clustering produces 322,620 multi-address clusters (295,978 at 2-hop; 290,319 at 3-hop). Among the 3,203,502 addresses in multi-address clusters, 3-hop clustering reduces the apparent number of distinct creators by roughly 11 times. Under 1-hop clustering, the median multi-address cluster holds 3 addresses and the largest holds 10,531.
- **Coin creation is highly concentrated.** Addresses in multi-address clusters account for 62.99% of all coins while being only 48.02% of creator addresses (1-hop). The top 1% of creator addresses alone create 38.89% of coins; after clustering, the top 1% of clusters create 52.99% (1-hop), 57.47% (2-hop), and 58.57% (3-hop).
- **Clusters survive adversarial checks.** In the top 1% most active clusters, the median cluster’s most active address creates only 26.33% of its coins and the top three create 50.33%; the median number of addresses needed to reach 90% of a cluster’s coins is 25.5. After worst-case single-address deletion, the largest remaining component retains a median of 80.4% of addresses and 89.4% of coin output.
- **Coordinated sells are common and multi-account.** DP1 (atomic, all senders sign one transaction) finds 4,402 dumps in the 1% sample and 3,510 in the 5-day sample, nearly 8,000 total. DP2 finds 9,270 dumps at a 5-second window, 66,706 at 1 hour, and 88,730 at 1 day. The median dump involves 7 senders under DP1 (max 7) and 7 under DP2 (max 17); the 5-day sample contains one dump with 312 senders.
- **Senders and dumpers are reused.** Graph analysis of the two largest DP1 clusters shows many senders repeatedly interacting with many dumpers across separate manipulations.
- **Dump proceeds are material.** DP1 dumpers sold 33,393 SOL; given SOL averaging roughly USD 160 and never below USD 80 during the period, the authors estimate USD 2.5M–5M in this 1% sample alone.
- **Copycats are abundant.** The strictest heuristic (shared name, symbol, description, and IPFS image hash) identifies 1,490,123 copycats against 647,986 originals; a name+symbol+image heuristic finds 1,866,178; a name+symbol-only heuristic finds 5,392,081. At minimum 1.5 million coins — over 10% of all coins — duplicate an original’s name, symbol, description, and image. Copycats more often omit the “pump” address suffix than originals (0.26 vs 0.14 under the strictest heuristic). 9,047 copycats caught only by the second heuristic carry a “Deployed using https://[Service-C]” description, pointing to automated deployment.
- **Social signals track coin creation.** 3.5 million coins (23.5% of all) are created after Twitter or Truth Social posts, and 31 individual posts each helped their respective coin creators earn at least 1 million USD.
- **Manipulation is sold as a service.** MMaaS websites and tools advertise multi-address creation, low-latency token copying, template- or AI-generated comments, and anti-detection features such as CAPTCHA solvers and proxy rotation, typically charging per-trade fees such as 1%. The authors cite the SEC’s “Market-Manipulation-as-a-Service” framing and note one existing court case.
- **Scale context.** pump.fun has grossed over 100 million USD in daily trading volume and nearly 90 billion USD total; its competitors (BONKfun, Raydium LaunchLab) together host about 1 million coins with a combined market cap of 7 billion USD. The most popular coin, “Fartcoin,” peaked at a 2.5 billion USD market cap. The platform saw a median of 17,926 new coins per day and a peak of 71,735 in one day.

## Methodology in Plain English

The authors start from a convenient chokepoint: pump.fun tracks every coin through a single on-chain “Token Mint Authority” address, so pulling all transactions involving that address yields every coin ever created on the platform. They filter to successfully parsed coin-creation transactions, obtaining 15,245,966 coins over 730 days (151,780,482 Solana slots), and record each coin’s creator as the address paying the first transaction’s fees. Because fetching every trade for every coin is cost-prohibitive, they analyze complete transaction histories for a random 1% of coins and, as a consistency check, for all coins created on five specific days (two chosen for extreme activity, three at random), excluding only a handful of coins with more than 5 million transactions each.

Manipulation detection uses paired heuristics: a strict version that only flags activity impossible to produce through the normal website (same-transaction buy/sell for wash trading; multiple signers funneling tokens to one seller atomically for dumps), and a looser version that allows a time gap and is therefore more inclusive but noisier. For creator obfuscation, they build a funding graph where each creator address points to the address that first sent it SOL or tokens, drop nodes labeled by Arkham as third-party services (exchanges, bridges, contracts), and merge addresses into clusters across one, two, or three funding hops — stopping at three because coverage saturates. They stress-test those clusters by deleting the single most connective address and measuring how much of the cluster survives. Copycat detection compares coin metadata and relies on IPFS URLs, which embed a file hash and therefore give exact image-duplicate matching (though not near-duplicate matching). Finally, they enrich on-chain data with pump.fun metadata, profile-page comments (collected for the 1% sample: 58,715 coins with comments, 694,483 comments), and daily price data for 154,513 graduated coins (57,623,422 data points), while respecting pump.fun’s Terms of Use with 4–5 second gaps between requests.

## Why This Matters

For research, the paper shows that manipulation on high-throughput, low-fee chains is best studied at the *event* level rather than the account level: detecting individual wash trades or dump transactions yields fewer false positives and reveals intent better than thresholding accounts that repeatedly buy and sell. It also sets a new scale bar for the meme-coin literature (all 15.2 million coins versus 0.04% samples in prior work) and adds creator obfuscation, post-driven coin creation, and MMaaS to the list of phenomena worth measuring.

Real-world applications:

- **Traders and wallet/chain-scanner developers** can flag coins whose volume is dominated by atomic same-transaction trades, or whose creator belongs to a large funding cluster, instead of showing raw volume and holder counts.
- **pump.fun and other launchpads** can price or restrict single-transaction buy-sell patterns, surface creator-cluster concentration, and deduplicate copycat deployments at creation time.
- **Software development platforms and app stores** can act on MMaaS tools that advertise multi-account creation, AI comment generation, CAPTCHA solving, and proxy rotation.
- **Regulators and law enforcement** gain measured prevalence figures and a description of the as-a-service layer that lowers the expertise barrier to market manipulation.

Industry relevance is direct: the paper documents an ecosystem where a platform with nearly 90 billion USD in cumulative trading volume has manipulation built into its economics — creator fees, graduation mechanics that reward visibility, and near-zero creation cost — and where third parties sell the manipulation tooling commercially.

## Future Directions

- **Capturing advanced wash trading.** The authors searched for the pattern where address A buys coins, sends them to B, and B sells, and found none, but they explicitly note they do not capture all advanced forms, such as buying, splitting, and selling coins separately — a detection gap worth closing.
- **Extending beyond a single launchpad.** With competitors (BONKfun, Raydium LaunchLab) hosting around 1 million coins combined, the question of whether factory-style manipulation migrates across platforms, and how detection transfers, remains open.
- **Evaluating mitigations.** The paper proposes interventions for five stakeholder groups but does not measure their effectiveness; testing whether flagging atomic wash trades or creator clusters actually reduces retail losses is the logical next experiment.
- **Understanding the MMaaS market and social pipelines.** The prevalence, pricing (e.g., 1% per-trade fees), customer base, and measurable returns of MMaaS tools — and how Twitter or Truth Social posts convert into creator earnings — are characterized qualitatively here and invite systematic study.

## Target Audience

Blockchain security and measurement researchers; exchange, wallet, and chain-analytics engineers building trade-signal or fraud-detection features; platform Trust & Safety and policy teams at token launchpads; financial regulators and enforcement units working on crypto market abuse; and technically literate traders who want to understand how volume, ownership, naming, and social signals are manufactured.

## Authors’ abstract

Once complex, creating and deploying a new cryptocurrency has become trivial. Coin launchpads now allow users to generate a new coin with merely a few clicks, at a minimal cost. Launchpad popularity has grown in tandem with the rise of "meme coins," which usually do not offer any novel technological properties and are purely created for fun. The most prominent coin launchpad, pump.fun, has gained significant traction, grossing over 100 million USD in daily trading volume.…

[The full abstract on arXiv](https://arxiv.org/abs/2609.10246)

[Read the original paper](https://arxiv.org/abs/2609.10246)

---

Canonical: https://crypto.world/paper/meme-coin-factories-uncovering-large-scale-manipulations-on-pump-fun-2609.10246
