pump.fun — Coordinated Wash-Trading Cluster Monitor
A live, on-chain detector that scores pump.fun launches for coordinated, risk-free round-trip ("wash") volume — disposable wallets buying and selling the same token with net position ≈ 0. It surfaces candidates consistent with fabricated volume, benchmarks them against a random control, and learns which throwaway round-tripper wallets recur. A triage and hypothesis tool — not an adjudication.
Correction (kept on the record). An earlier version treated a recurring set of ~16 wallets as an "enterprise fingerprint." On-chain verification of pump.fun's global config showed 15 of those 16 are the platform's own protocol fee-recipient wallets — they touch every pump.fun coin by design, so they are not a cabal and are now excluded from the signal entirely. The wash-trading measurement below is independent of them and unaffected.
connecting to scanner…
RICO is used here only as an analytical lens, not a conclusion. The Racketeer Influenced and Corrupt Organizations Act (18 U.S.C. §§ 1961–1968) supplies a vocabulary — enterprise (§1961(4)), pattern (§1961(5)), predicate acts such as wire fraud (§1343) and money laundering (§1956), and the operation-or-management test of Reves v. Ernst & Young. On-chain data can show coordinated, risk-free round-tripping and recurring wallet structure; it cannot, by itself, establish common beneficial control, fraudulent intent (scienter), reliance, proximate cause (Anza/Hemi), or that any named person or company is a member of a legal "enterprise." Those are questions for counsel and a court, not for a heuristic.
What this establishes — and what it does not.
- Establishes: specific disposable wallets generated buy/sell volume on the same mint with net position ≈ 0 above a 1-SOL floor within the sampled window — i.e. fictitious volume in the economic sense. Some of these throwaway round-trippers reappear across otherwise-disjoint launches (pump.fun's own fee-recipient wallets are excluded from this count).
- Does NOT establish common control. Shared funding/timing is consistent with one operator or with shared bundler/SaaS bot infrastructure sold to many customers, MEV/arbitrage bots, or market-maker churn. No controller-clustering error rate has been validated here.
- MEV / arbitrage not fully excluded. Same-block round-tripping also arises from sandwich bots and bonding-curve↔AMM arbitrage; residual wash net of those is not separately reported yet.
- Selection bias. The "core" set is defined by the signature under test, so a high in-set wash ratio is partly tautological. Platform-wide base rates (pump.fun graduation historically ≈0.5–1.8%) are the proper denominator and are not yet controlled for.
- The "organic" control is illustrative (n=1), not a matched cohort. Graduation deltas are confounded by KOL marketing, memetics, timing and luck.
- Exchange labels are heuristic terminals. A major-CEX tag (from Solscan) marks a KYC boundary where the on-chain trail ends; deposits there are consistent with any user off-ramping and are not alleged to be enterprise members. The specific venue is intentionally not named on this page.
- Fee math is arithmetic, not intent. If X% of a coin's volume is wash, X% of the protocol fee it generates is wash-derived — this says nothing about the platform's knowledge or intent. pump.fun fees are now dynamic (≈0.05–0.95% by cap), so a flat 1% model is an approximation.
Live forward scanner — new wash clusters & recurring round-tripper wallets
The scanner scores recent launches on wash ratio + count of fresh coordinated big buyers (0–1). Any disposable wallet it sees round-tripping at ≥2 distinct coins is learned as a repeat offender (pump.fun fee recipients are excluded). The learned set grows only from genuine repeat round-trippers.
—
candidate clusters detected
—
recurring wallets discovered (≥2 clusters)
—
wallets under observation
—
learned repeat round-trippers
Base rate / negative class — does the signature fire on random launches?
The sharpest objection to any cluster detector is circularity: if you only look at coins you already flagged, of course they look flagged. So each cycle the scanner also scores a random sample of recent pump.fun launches that did not come from the fingerprint — the negative class. If the signature were noise, random launches would score as high as the flagged ones. They don't.
—
random launches scored (control)
—
median score of random launches
—
of random launches score ≥ 0.5
—
of random launches touch a learned round-tripper
Side by side — flagged clusters vs random launches
| metric | flagged clusters | random launches (control) |
|---|
Candidate clusters (live) — wash % is an upper bound (MEV/arbitrage not yet separately excluded)
| token | score | repeat hits | fresh big buyers | wash % (≤) | gross SOL | first seen |
|---|
| waiting for first scan cycle… |
Newly-discovered repeat round-tripper wallets
| wallet | type | # clusters | last seen |
|---|
| none yet — promoted after a wallet recurs across ≥2 clusters |
Scanner log
starting…
Frozen case study — the originating cluster
The signature was derived from one cluster captured in a frozen, hash-pinned snapshot (so cited figures don't drift). Snapshot id: —. These are point estimates from the sampled windows, not validated against ground truth.
—
aggregate wash ratio (round-trip ÷ buys)
—
gross volume measured (consistent w/ fabricated)
—
coins in case-study sweep
—
graduation rate (vs ~0.5–1.8% platform base rate)
Two exhibits (run the machine hardest)
SPCX
—
of buy volume round-tripped by the same wallets
IRAN
—
of buy volume round-tripped by the same wallets
In both, coordinated buyers each put in ~45–75 SOL and sold ~the same back (130+ buys / 130+ sells each, net ≈ 0). Capital shells are fresh per launch (0 overlap); the disposable round-trippers are what recur across coins — the learned signal (pump.fun fee recipients excluded).
Funding topology (CEX = labeled terminal, not a network node)
Fee attribution (arithmetic only)
| bucket (case-study sweep) | SOL | USD |
|---|
| Gross traded volume (buy+sell) | | |
| — round-tripped wash volume | | |
| Solana base fees | | |
| Jito MEV tips | | |
| pump.fun/PumpSwap protocol fee (≈1% model) | | |
| — fraction attributable to wash volume | | |
The open question, posed neutrally: a platform fee is a cut of volume. On these specific coins, a large share of that volume is wallets transacting with themselves. What fraction of fee revenue on flagged coins is wash-derived, and what (if anything) follows from that, is for the reader and for counsel to weigh — this tool only measures the volume.
Method & limits. Public Solana ledger via Helius RPC + enhanced-tx API; market data Dexscreener/GeckoTerminal; exchange labels Solscan (heuristic). "Wash" = SOL round-tripped by wallets that both bought and sold the same mint above 1 SOL within the sampled window (a conservative floor). Forward detections are candidates from a behavioural heuristic with an unvalidated false-positive rate; MEV/arb/market-making are not yet separately excluded. The case-study snapshot is frozen and hash-pinned; live scanner figures update each cycle. Wallets are pseudonymous; no real-world identity, and no named person or company, is asserted to be a participant. This is on-chain analysis and protected commentary — not legal advice, not an accusation, and nothing here has been adjudicated. RICO terms appear solely as an analytical framework for the reader to weigh.
Real forensic chain analysis — not larping. · I'm not vengeful; I just wish justice for all. ·
freestacc.io