Exact definitions, every threshold, and what this method cannot see. Thresholds below are read from the running configuration.
Ingestion
Every pump.fun create (instructions create and create_v2 of program 6EF8rrecthR5Dkzon8Nwu78hRvfCKubJ14M5uBEwF6P) uses the program's mint-authority account TSLvdd1pWpHVjahSpsvCXUbgwsL3JAcvokwaKt1eokM. One Durable Object holds a Helius WebSocket subscription to transactions that mention it (pushed within a second or two), and every 10 seconds lists that account's signatures from a durable cursor with getSignaturesForAddress, so a create the socket missed is still found, from the chain. A transaction counts as a launch when it carries pump.fun's CreateEvent; PumpSwap pool creation uses the same account and is skipped.
PumpPortal's free new-token stream was the first choice. Measured against the chain it delivered 33.5% of creates (448 of 1336, 60 minutes on 2026-10-26), so it is kept connected only to keep measuring it.
Slot 0 and the early slots
Slot 0 is the slot of the create transaction. Slots +1 and +2 are the early slots. Each is read with getBlock (transactionDetails: "full", maxSupportedTransactionVersion: 1: mainnet carries v1 transactions, and with 0 the call fails for any block that holds one). A buy is a pump.fun TradeEvent with is_buy for the mint, so buys through any router, bot or bundler program count; one row per buyer per transaction. A failed transaction that lists the mint and the pump.fun program is a failed attempt, kept apart. For every buy: buyer, quote paid (SOL, USDC or a pump coin), tokens, position in the block, fee payer, compute-unit price, and any SOL sent to a Jito tip account.
Shares are of the token's total supply from its CreateEvent (1,000,000,000 tokens on pump.fun), computed in integers.
Classes (first match wins)
creator The creator wallet (the creator of the CreateEvent). Its buy inside the create transaction is the dev initial buy, a normal pump.fun feature, reported apart from everything else. pump.fun lets a signer create a coin for another creator address; such a signer is classified like any other buyer and marked.
linked One concrete on-chain link to the creator, shown with the transaction that proves it:
a. a direct SOL transfer between the buyer and the creator within 7 days before the buy. The brief's rule is creator → buyer; Block Zero also counts buyer → creator, and the proof says which way the SOL went.
b. funded by a wallet that also funded the creator, both within 7 days before launch. Exchange hot wallets are never a common funder (list).
c. sent its tokens to the creator, or to another LINKED wallet, within 24 hours after launch.
cluster Not linked to the creator, but at least 3 slot-0 buyers were funded by the same non-exchange wallet within 7 days. Shown as "N slot-0 buyers funded by the same wallet".
unlinked Everyone else in slot 0 (typically snipers).
Funding is a native SOL transfer (System Program transfer, transferWithSeed, or createAccount owned by the System Program) whose source signed the transaction, of at least 0.01 SOL. A program paying out from its own account cannot sign and is never funding; the floor keeps address-poisoning dust from looking like a shared funder. A transfer in the create slot counts only when it comes before the buyer's own buy in the block.
Early-slot buyers get the CREATOR and LINKED checks that need no history of their own (rules a and b from the creator's side, and rule c). Their own funding is not read, so they are never CLUSTER.
Labels
CLEAN
No LINKED or CLUSTER buy in slot 0, and the dev initial buy is at most 5% of supply.
DEV-HEAVY
Dev initial buy above 5% of supply, no LINKED or CLUSTER buy.
LINKED BUYS
LINKED + CLUSTER wallets in slot 0: at least 2 wallets, or at least 3% of supply.
SNIPED
UNLINKED slot-0 buyers took at least 10% of supply. A launch can be LINKED BUYS and SNIPED, or CLEAN and SNIPED (snipers are not linked to the creator).
QUIET
No buy at all in slot 0 besides the creator's.
One linked wallet under 3% gets no label, and the record says why. A partial record (a history that could not be read, a work cap reached) never gets CLEAN or DEV-HEAVY, because both say something was absent; what was found is still shown.
Dev record
Daily cards name the cleanest dev of the day: at least 2 launches that day, every one CLEAN, ranked by how many of them other wallets bought in slot 0. The weekly list on the home page takes devs with at least 3 launches in 7 days, every one CLEAN, ranked by launches still trading after 24 hours.
A dev is the creator wallet, plus the wallets that appear as LINKED in two or more of its launches (shown as a second group, each appearance with its proof). For every dev: launches seen, % CLEAN, % LINKED BUYS, the average LINKED share of supply, the average share the creator sold in the first hour, and how many launches still trade after 24 hours (24-hour volume above $1,000 on DexScreener). History starts when the indexer started; looking up a wallet also reads its own last 30 days of creates (at most 2,000 transactions and 50 launches), labeled backfilled. The Clean Launch badge needs 3 or more analyzed launches, every one CLEAN.
First hour and 24 hours
At +1 hour, each slot-0 and early-slot buyer's token account is read for the first hour: sold (pump.fun sells, or a balance drop in a transaction that paid the wallet SOL), moved (a plain token transfer to another owner with no trade in the same transaction) or held, as shares of what the wallet bought in slots 0 to +2. Also: the share of supply in the 10 largest token accounts (the bonding curve's own account excluded; accounts, not people), and the market cap from the bonding curve's reserves (price = virtual quote reserves / virtual token reserves, times supply, in USD at the quote's price). At +24 hours rule c is checked over the full window and DexScreener's 24-hour volume is read.
Exchange hot wallets
Withdrawals from a large exchange fund thousands of unrelated wallets, so these addresses never link two buyers. The list, with its sources, is in the repository (data/cex/):
Wallets funded through exchange withdrawals cannot be linked to each other or to the creator: the exchange breaks the trail. Bridges, swap services and other intermediaries that are not on the list can, rarely, put unrelated wallets in one CLUSTER; the funder is always shown.
Only one hop: a wallet funded by a wallet that was funded by the creator is not linked.
A Jito bundle (or any shared block position, tip or fee payer) is shown, but is not proof of common ownership by itself and never links anyone.
Busy wallets: funding history is read newest first, at most 100 transactions per wallet; for a wallet with more in the window the record says how far back it was read. A busy wallet is read again at most once per 60 minutes.
Per launch, the funding of the 15 largest slot-0 buyers is read and the 4 most recent non-exchange funders of the creator are followed; past that the record is partial and says so.
The daily RPC budget (150,000 credits) bounds the work; over it, launches are recorded partial instead of guessed.
A pump.fun create can name a creator other than its signer. When the creator's address is not in the create transaction, a creator's own history does not show that launch, so the backfill can miss it (live indexing does not).
Per-buyer detail is kept 30 days; summary rows, the strip and LINKED/CLUSTER rows with their proofs are kept for good.
Token names are what creators typed; a short list of slurs is masked where Block Zero displays them. The API returns them unchanged.
Coverage
Last 24 hours: 90.76% of on-chain creates analyzed (2,023 of 2,229, the reference being every successful mint-authority transaction counted from chain each hour, pool creations excluded). The WebSocket alone delivered 98.8%; PumpPortal's stream 36.4%.
No uncovered period recorded.
Test vectors
Five synthetic slot-0 blocks with their expected classifications. They run through the same code as live launches, here, in the test suite and in demo mode. Addresses are synthetic and are never linked to an explorer.
Vector A: Dev buy only ✓ computed = expected
The creator buys 8.0% inside the create transaction; nobody else buys in slot 0; two wallets buy in slot +1.
Six unrelated wallets buy 11.3% right behind a 1.5% dev buy and three attempts fail. The same exchange hot wallet funded the creator and one sniper (excluded), and two snipers share a funder (two is below the cluster minimum of three).
#
wallet
bought
class
why
40
Creator
1.5%
creator
creator wallet
41
Sniperone
2.0%
unlinked
no link found
42
SniperTwo
2.1%
unlinked
no link found
43
SniperThree
1.8%
unlinked
Funded from an exchange wallet (Coinbase) 24 h before launch: not counted as a link
Vector D: Rules b and c, and a sender to the creator ✓ computed = expected
A hub funds the creator and two buyers (rule b); a third buyer later sends its tokens to the creator and a fourth to a linked wallet (rule c); the wallet that sent the creator SOL six hours earlier also buys (rule a, sent).
#
wallet
bought
class
why
77
Creator
3.0%
creator
creator wallet
78
MainWaet
4.0%
linked·a
Sent 5.00 SOL to the creator 6 h before launch
79
HubKidone
3.0%
linked·b
Funded by the same wallet as the creator: 1.20 SOL 2 h before launch (the creator got 2.00 SOL 3 h before launch)
80
HubKidTwo
3.0%
linked·b
Funded by the same wallet as the creator: 1.20 SOL 2 h before launch (the creator got 2.00 SOL 3 h before launch)
81
Taterone
2.5%
linked·c
Sent its tokens to the creator 40 min after launch
82
TaterTwo
2.5%
linked·c
Sent its tokens to a linked wallet 50 min after launch
Vector E: Cluster and snipers ✓ computed = expected
Four buyers funded by the same wallet (not linked to the creator) take 9.0%; five snipers take 12.5%, three of them withdrawn from the same exchange (excluded, so no cluster); five attempts fail.
#
wallet
bought
class
why
10
Creator
0.5%
creator
creator wallet
11
Groupone
2.4%
cluster
One of 4 slot-0 buyers funded by the same wallet: 1.00 SOL 50 min before launch
12
GroupTwo
2.3%
cluster
One of 4 slot-0 buyers funded by the same wallet: 1.00 SOL 45 min before launch
13
GroupThree
2.2%
cluster
One of 4 slot-0 buyers funded by the same wallet: 1.00 SOL 40 min before launch
14
GroupFour
2.1%
cluster
One of 4 slot-0 buyers funded by the same wallet: 1.00 SOL 30 min before launch
15
Snipeone
2.5%
unlinked
Funded from an exchange wallet (Binance) 3 h before launch: not counted as a link
16
SnipeTwo
2.5%
unlinked
Funded from an exchange wallet (Binance) 4 h before launch: not counted as a link
17
SnipeThree
2.5%
unlinked
Funded from an exchange wallet (Binance) 5 h before launch: not counted as a link