method
How this is measured,
and where it would lie to you.
The measurement is simple enough to state in a sentence and has at least several ways of going confidently wrong. Seven of them were found by producing a wrong number first — including one where this site fell for its own subject. They are all below, with the wrong numbers left in.
The claim
One edge that cannot be avoided
A Solana keypair costs nothing to generate and can do nothing until somebody sends it lamports. Creating a pump.fun coin costs rent and fees, so the wallet that mints must have been funded first, by an account that already had money. Metadata can be blanked, an IPFS host swapped, a launcher's advertisement deleted from the description — but that payment happened on chain and stays there. OneHop reads it.
The pipeline
- Record. Walk pump.fun's creation feed and keep every launch: mint, creator wallet, timestamp, metadata URL.
- Trace. For each distinct creator wallet, read its whole signature history from a fast endpoint. Its oldest transaction is a candidate genesis.
- Attribute. In that transaction, find who paid — by balance delta, not by instruction shape, because funding arrives in more shapes than
system::transfer. - Confirm. Ask an archival node whether anything exists before that transaction. If it does, the candidate was a retention artefact and the wallet is dropped from every claim.
- Characterise. For each payer: how many wallets it seeded, how many distinct amounts it used, what it holds now.
- Test. Ask whether its repetition is surprising at all, given what amounts everybody else sends.
Steps 3 and 4 are shared code, not a described procedure: js/trace-core.js is imported
by the scanner and by the browser, so the Trace and Re-run
pages cannot reach a different verdict than the published snapshot did.
The traps
Ways this measurement lies
Each of these produced a number, and seven of the numbers were wrong. Five were more dramatic than the truth, which is the reliable tell. One went the other way and nearly deleted a real finding, which is the reminder that the tell is a heuristic and not a law.
The null model
Why “they all sent the same amount” is not enough
A payer that made two payments of 0.1 SOL has a perfect coherence of 1.0, and 0.1 SOL is the most common amount in the whole dataset because it is what a person types. Counting that as one operator is how the naive figure gets to —.
So every payer is tested against the distribution of amounts everybody else sends: given k payments, how often would m of them happen to carry the same value? Its own payments are removed from the pool first — otherwise a payer with six identical payments partly defines the distribution it is being graded against. The resulting p-values are corrected across all — payers tested, because two hundred tests at p ≤ 0.05 manufacture ten false positives by construction.
Deliberately conservative in one direction
An amount that many unrelated payers use is treated by this test as unsurprising — even though shared use across unrelated payers is exactly what shared launcher software looks like. That is a real cost and it is paid on purpose: folding it in would let the site's most striking observation grade its own homework. It stays on the findings page as a description, feeding nothing.
What it refuses to do
Naming nobody
The largest payers here hold hundreds of thousands of SOL and fund thousands of strangers. They are obviously exchanges, and this site still does not say which. Press coverage once put a famous wallet's share of a token at ~60% where the chain said ~9%; attribution is not identity, and a measurement site that starts labelling addresses by reputation has stopped measuring. Every payer is described by what it does — how many wallets, what amounts, what balance — and linked to a block explorer so you can decide for yourself.
The same applies in the other direction. A wallet classified as one hand means: this account paid several minting wallets an identical, computed amount, and that pattern is unlikely to be chance. It does not mean fraud, it does not identify a person, and plenty of legitimate tooling funds wallets programmatically.
Reproducing it