Method
How each figure is produced
Everything here is read from the Solana chain. Where we recorded a launch's creation transaction, its record cites it and you can decode it yourself rather than take our figures on trust; where we did not, the record says so and the figures rest on our observation at the time. This page states what is recorded, how the one classification we apply is defined, what was done to test it, and, the part that matters most, what we refuse to say.
What is recorded, and when
A collector decodes the launch program's own events as they happen, on pump.fun and Raydium LaunchLab, and a chain scanner reads Meteora DBC launches out of whole blocks. For every launch we write down: the creator, the share of supply the creator took in the creation transaction, what the token claimed to be, the state of its bonding curve as it moved, and whether that curve completed. These are facts about a moment. They stop being observable once the float is spread across wallets, which is why they are recorded live rather than inferred later.
Two further facts need the venue's events to name the wallet that traded, and not every venue's do: every distinct wallet that bought on the bonding curve, and whether the creator sold. On pump.fun the trade events carry the trader, so both are recorded. Where they do not, those fields are empty rather than zero, and a launch there is never described as having had no outside buyer. An absence of data is not a finding, least of all one that accuses.
Coverage begins 2026-09-02 12:29 UTC, with 46 minutes of recorded downtime. A launch that happened while the collector was down has no record, and is reported as unobserved rather than as anything else.
What "checked, no markers found" means
It means we watched the launch, checked it against every pattern below, and found none of them. It is a statement about what we checked, not a judgement about the token: not a prediction, not a recommendation, and not a claim that it will hold its value. A record is described that way only when every one of these is true of it:
| Test | Threshold | Why |
|---|---|---|
| Creator's share in the first block | under 20% | above this share, the creator holds more of the supply than everyone who buys on the curve combined |
| Distinct outside buyers on the curve | at least 30 | below this, the curve was completed by a handful of wallets rather than many |
| Time to complete the curve | over 60s | a curve filled this fast was completed before other wallets recorded a buy on it |
| Largest single buy on the curve | under 40 SOL | one buy that completes a curve is a purchase of the float, not a market |
| Creator sold | no | self-explanatory |
Liquidity is not one of these tests, and until 2026-09-09 it was. Every test above is a fact about the first blocks of a token's life: once true, always true, and not readable from the token's present state once the float has been spread. A pool balance is a reading taken at one moment and it decays. Requiring both before a record could read no markers found meant an hour of unanswered RPC calls silently withdrew findings about the past: over seven days, 423 launches matched every line above and ten were published. We still read the pool, still refuse to quote a balance we could not confirm, and now show it beside the launch record with the age of the reading instead of gating the record on it. A row reading not read is a gap in our pool coverage, never a finding about the token.
How a linkage is established
Where this register says two creator wallets are linked, the claim rests on published thresholds, applied identically to every launch, checkable in the downloadable record. "Linked" is never our say-so.
| The artifacts | Three instruments: a byte-identical launch image (matched by sha256), a byte-identical metadata document (sha256), and an identical ticker family - the exact symbol and name pair. Similarity never counts; only identical bytes or identical pairs. |
| The template rule | An artifact shared by more than 20 distinct creator wallets is excluded before anything is counted - a platform default or a famous meme is not a fingerprint. The ceiling is recorded in the file's own metadata beside the counts it produced. |
| One coincidence is not an edge | A pair of wallets is linked only on two or more distinct shared artifacts, or two different instrument kinds. One shared artifact is recorded as a lead and produces no edge. Each edge cites every artifact behind it, by hash, in the record. |
| Where the instrument stops | A connected group larger than 100 wallets gets no group label: at that size transitive linking stops resolving individual hands, and the exclusion is stated in the file rather than the group being published as one. Measured when the rule was set: the largest genuine group held 30 wallets; the excluded component held over 12,000. |
| Corroboration is a count | Each published group carries how many independent instruments place its members together: the artifact kinds above, a shared funding wallet, and first fundings within one hour of each other. A count, never a grade. |
| Shared infrastructure clears, not condemns | Where the shared funder is infrastructure-profile - its outflows enumerate at hundreds of transactions an hour, or its funded wallets each trade mostly different launches - the group is recorded as cleared by that instrument, with the funder and the measurements cited. A payment service funds strangers the way an operation funds its own wallets; saying so is part of the method. |
Every threshold above is applied by the same published code on every build, and every count it produces sits in the record's documented tables next to the exclusions it made.
What we refuse to say
The record declines to answer in four situations, and says which:
| Situation | What you get |
|---|---|
| The launch happened before coverage, or while the collector was down | UNKNOWN, with an offer to rebuild the record from chain history |
| A rebuild could not read every transaction | UNKNOWN, because a truncated history looks exactly like a quiet launch |
| The pool balance could not be read | no liquidity figure quoted, and the row says not read. The launch record is unaffected |
| We hold no record and the address has no bonding curve we can rebuild from | we say so, rather than guess |
Rebuilt records
A launch we did not watch can often be reconstructed: a bonding curve is a single account whose whole transaction history is readable, so the same on-chain events can be decoded later. Those pages are marked rebuilt. The figures are the same events read afterwards, and are checked the same way, but a rebuild cannot tell you what a token claimed to be at launch, because the name, image and links live off-chain behind a URI the operator can repoint. That, and only that, is genuinely unrecoverable.
What we count as a launch
Most tokens created on Solana are not launches. Liquidity pools mint pool tokens, concentrated liquidity mints a position NFT for every position, prediction markets mint an outcome token per outcome, and at least one program mints tokens with no supply at all. Over one measured window, 155.7 token creations a minute contained about 30.8 that were launches. So a count of creations and a count of launches are different numbers and we publish them as different numbers.
Deciding which is which is a judgement, and it is ours rather than the chain's. When a launch is recorded through a venue's own events, the venue declares what it is. When it is found by reading blocks, nothing declares anything and we apply a test. This is that test, stated so it can be disagreed with:
| A mint is recorded as looking like a launch when | threshold |
|---|---|
| tokens were actually minted in the creation transaction | supply > 0 |
| it is divisible like a currency rather than counted like an NFT | at least 6 decimals |
How well it does, measured rather than asserted. Over 392 blocks and 433 token creations, labelled by the program that made them, the test keeps 97 of 109 known launches and admits 6 of 317 known non-launches. The dozen it misses are mostly mints whose tokens are created in a later transaction, which is a real pattern and not a non-launch. It is therefore stored as a column and never used as a filter: every token creation is recorded either way, so a reader who disagrees with the threshold can recompute from the same file rather than asking us what we discarded. A mint failing this test is one that does not look like a launch to us. It is not a statement that it is not one.
Where we have identified the program that created a mint, that is a stronger answer than the test and does not depend on our judgement at all. Those programs are listed on the launch programs page with the source for each, and the list grows as programs are identified.
Known limits
Stated because a method page that lists no weaknesses is marketing.
| The labelled set is drawn from this archive, so it cannot contain a factory that uses a fresh ticker every time. It is a precision test, not a census. |
| Thresholds are judgements. They are set where no launch in the labelled set read as no markers found, not where some theory says they belong. |
| Operator attribution describes wallets' behaviour inside this archive only, and says nothing about intent or identity. |
| A trade is timestamped when we decode it, not by block time, so the interval between a launch and the buy that completed its curve is only as fine as the batch both arrived in. Where that interval reads as zero we say the events arrived together, rather than quoting a duration. The slot is published in trades for anyone who wants to settle it exactly. |
| Each venue has its own coverage and a launch is answered against its own venue's windows: pump.fun from 2026-09-02 12:29 UTC, Raydium LaunchLab from 2026-09-12 02:10 UTC, Meteora DBC from 2026-09-14 10:55 UTC. Launch venues other than pump.fun, Raydium LaunchLab and Meteora DBC are not recorded at all, and a launch on one of them reads as unwatched rather than as clean. |
| Counts of distinct buyers, and whether the creator sold, are recorded only on pump.fun, whose trade events name the trading wallet. Elsewhere those fields are empty, and every finding below that counts buyers is measured over pump.fun alone. |