How this is measured

1. Every figure here is a floor, and one of them is exact

A creator's earnings are reported as a lower bound: what is sitting in their fee vaults right now, with locked rent excluded. Not a lifetime total. The real figure is this one or higher, never lower.

That is not caution, it is the strongest claim the chain will support without a paid archival node, and the limit was measured rather than assumed. The public RPC retains roughly 2.2 days of ledger — getFirstAvailableBlock sits about 473,000 slots behind the tip — so the transaction that funded a vault is usually long gone. An earlier version of this site tried to reconstruct lifetime intake from transaction history and produced vaults visibly holding millions of lamports with no deposit anywhere in their history. It was caught by the reconciliation check below and never published.

The bound is rigorous in both directions, and that is what makes it usable:

• No vault means exactly zero — this half is proven. pump.fun creates the fee vault on the first fee that accrues, and a claim can never drain it below the rent-exempt minimum. Two of the largest coins in the sector have bonding-curve vaults sitting at exactly 890,880 lamports, which is that minimum to the lamport. So a vault that ever received a fee still exists, and its absence is evidence rather than a gap.
• A vault that exists proves only a minimum. Its balance is what has not yet been claimed. A creator who withdrew last week reads as poorer than they are, and this measurement cannot see the difference.

Most creators are in the first case, so the median is exact while the top of the distribution is a floor — the figure the site leads with is the one the instrument measures best. Note also which way the error runs: under-measuring the top understates inequality. It cannot manufacture this site's own conclusion.

3. What was tried first, and why it was thrown away

The obvious method is to reconstruct each vault's lifetime intake from its transaction history — the sum of positive balance deltas across every transaction that ever touched it — and check it against an accounting identity: received − claimed must equal the current balance.

That was built, and the identity is what killed it. It held on 6% of rows. The cause was not a coding error but the ledger retention above: vaults holding millions of lamports had no deposit anywhere in their visible history, because the deposit happened more than two days ago. Three consecutive scans were refused publication by the validation gate before the cause was found.

The check earned its keep by failing. A version of this site that shipped those numbers would have reported confident lifetime totals that were missing most of their history — and reported them as exact. What replaced it claims less and can support what it claims.

Rows whose RPC calls fail outright are still marked unknown-source-unreachable and excluded from every denominator. A source that failed is not a source that said "no": in a sibling project a single transient timeout briefly became a finding, and this is the guard against repeating it.

5. What this cannot see

Stated plainly, because a measurement that hides its blind spots is worth less than one that names them.

• Fees already claimed. The largest blind spot, and the one that sets the direction of every other error. A vault's balance is what has not yet been withdrawn, and the public RPC keeps only ~2.2 days of ledger, so a creator who claimed last month is indistinguishable from one who never earned. Active claimers — which means the successful ones — read as poorer than they are.
• Fee-sharing splits. pump.fun lets a creator route fees to several wallets. Income directed elsewhere is not counted against the launching wallet, so a team splitting revenue will read as several smaller creators rather than one larger one.
• One creator, several wallets. Nothing on-chain links two addresses belonging to the same person. A prolific launcher using a fresh wallet each time appears as many creators.
• Secondary income. Creators frequently hold their own token. Trading gains, airdrops and OTC sales are real income and are not fee income; none of it is measured here. This site measures fees, which is narrower than earnings.
• Creators who quit before the sample window. The pump.fun listing endpoint stops paginating after roughly 1,500 coins, so the sample is drawn from recent launches. Someone who launched a year ago and never launched again cannot be selected.
• Non-SOL denominated pools. Fee vaults are measured in SOL and WSOL. A coin paired against a different quote asset would accrue fees this scan does not read.

Every one of those biases the measured figure downward, and the site's argument is that the figure is low. That is the direction a careful reader should be suspicious of, so it is stated here rather than buried — and it is why nothing on this site is presented as a lifetime total. What can be said without qualification is the shape: a mean two orders of magnitude above the median is not an artefact of a floor, because the floor applies to both.