Methodology

Leverminal is a data platform; a swap made from a coin's page executes on Jupiter. Leverminal's fee is added on top of Jupiter's route, collected in the same transaction through Jupiter's integrator fee, and shown in the quote before you sign: 1%, or less for a wallet holding LEVERCAT. Jupiter, the pools, the network and each coin's transfer tax are separate and are not ours.

Leverminal's policy is to spend those fees buying LEVERCAT and burning it, and its operators may hold LEVERCAT. That is an interest, stated here rather than hidden: LEVERCAT's page, its holder rates and Edge all involve a coin Leverminal benefits from. The Terms say this at length.

The holders' tier. A wallet holding 100,000 LEVERCAT or more sees two things everyone else does not: coins in their first 24 hours, and the whale tracker. Without Edge, a coin under a day old is left off everything, not just greyed out: the board and its counts, search, the Hot strip, the launch ticker, alerts, the coin's own page and its evidence, a creator's list of launches, and the leading coins under a new pair. The board says how many it left off. Every other part of Leverminal, trading included, is the same for everyone.

To unlock it, the wallet signs a plain message naming this site and a one-time code. It is not a transaction and cannot move anything. Leverminal checks the signature, then keeps a signed cookie on this browser for seven days saying which wallet was proven, and nothing else. Whether that wallet holds enough is read from the chain on every request that needs it, from a one-minute cache: buying above the line starts Edge within about a minute without signing again, and selling below it ends Edge as quickly. When the chain cannot be read, the last balance seen within the hour decides; with none, the answer is no.

Where it goes wrong. The balance is what counts, not a lock: the same 100,000 LEVERCAT can be sent between wallets, and each can sign in while it holds them. Several people can share one wallet's session by sharing the wallet. A coin without a readable launch date is not treated as new. And a coin in its first day can still be found and traded elsewhere; Edge is what Leverminal shows, not what exists.

A team, or anyone, can pay for a coin's page to carry a cover image, a logo, a longer description and links: $99 in SOL or USDC, or $49 in LEVERCAT. Leverminal reviews every one before it shows, and a profile that is shown says it was paid for. The words and images are the payer's, not Leverminal's; they are shown as plain text and redrawn images, and links must be https and, for X, Telegram and Discord, point at those sites.

A boosted profile is cosmetic. It does not change the score or any check on this page, and it does not move a coin up the board. Payments are final, including when a profile is not approved. LEVERCAT paid for a profile is kept in Leverminal's wallet, not burned.

Where it goes wrong. Paying for a profile says someone wanted the coin to look cared for, which is not the same as the coin being cared for. A polished page over a rug is still a rug.

A coin's mint account points at a metadata file, and that file carries a description written at launch. StonkFun's launch form asks for one and StonkFun shows it nowhere. Leverminal reads the file on its server, never from your browser, keeps only the description, and shows it under the coin's name as plain text: a link in it is not made clickable. A creator who wrote nothing gets StonkFun's generated line, "<name> was launched on StonkFun."; those are StonkFun's words, not the creator's, and are not shown. Nor is a description made only of punctuation. A coin with a boosted profile shows the profile's reviewed description instead, and this one is not repeated. The file is read again every six hours.

Where it goes wrong. It is whatever the creator typed, and nobody checked it: it can promise anything. Some launch tools write their own line in it ("Deployed using …"), which is shown as written, because that is what the file says. The file sits wherever the creator's launch tool put it; a slow or refusing host means nothing is shown, which does not mean the creator wrote nothing. Legacy launches keep their metadata in a separate account that is not read, so they never show one; in a sample of twenty, every one carried StonkFun's generated line anyway.

One number out of 100 on a token page, sitting above the facts it is made of. Seven components, each a slide between two anchors and clamped at both ends, worth 100 points between them:

• Liquidity, 20 — the pool as a share of market cap: nothing at 0%, full marks at 10% and above.
• Holder spread, 20 — supply in the ten largest wallets: full marks at 20% or less, nothing at 60% or more.
• Holders, 10 — 2 points at 100 wallets, 6 at 1,000, 10 at 5,000 and above, log-linear between.
• Time on market, 15 — nothing under a day, 8 at a week, 15 at a month and beyond, log-linear between.
• Trading, 10 — a day's volume as a share of market cap: 5 points at 5%, 10 at 50% and above.
• Contract, 15 — full marks with the mint authority revoked, no freeze authority and no hostile extension; 7 off for each live authority, 8 off for each extension, floored at zero.
• The open, 10 — full marks for no bundle in the first second; 6 while the opening bundle holds under 5% of supply, down to nothing at 10%; nothing at all when the creator sold within a minute of the first trade.

A component nobody answered for scores nothing and is marked unknown, and the score reads partial with the count of them. Nothing is guessed and no coin is docked for a source being quiet, which does mean a partial score is a low score — so the count sits beside the number, and "how it adds up" opens every component with its points and what was measured.

Where it goes wrong. The inputs come from Jupiter, the mint account and the LaunchLab tape, and each can be stale or missing: Jupiter's budget is shared with live trades, the mint account may not have been read, a coin minutes old is not indexed yet, and a tape too long to walk back leaves the opening unknown. The same coin can score differently ten minutes apart with nothing having happened to it. The components and their weights are judgement, not something measured against outcomes. And a score cannot see intent: a coin can pass every one of these tests and still be somebody waiting to sell. It summarises the facts below it; it is not a safety rating, and it is not advice.

Every ticker is reduced to a key: Unicode compatibility form, invisible characters removed (zero-width, bidi and format characters), look-alike letters folded to Latin (Cyrillic and Greek letters, currency signs such as ₿), accents stripped, punctuation dropped, lower-cased. A leading $ is dropped, since that is how people write a ticker. Coins with the same key form a cluster. A ticker made only of symbols or emoji has no key and is clustered on its image address instead.

The main coin of a cluster is the largest by market cap; the first is the earliest created. A coin is marked “not the main X” when another coin in its cluster is at least five times its size and worth at least $10,000; the board and the token page apply the same rule. A coin with no market cap on record is never marked. The board hides a marked coin by default only when its own market cap is under $10,000, and leaves it out of the headline coin count. A search, your watchlist, the launch ticker and graduation alerts always show every coin. On a token page the cluster is searched for with the folded key, on StonkFun and, when Jupiter's budget allows, on other launchpads; the table lists the 40 largest and always includes the coin you are viewing.

Where it goes wrong. Common words collide: two unrelated coins called INU or BTC are not copies of each other, which is why the label says “not the main”, a statement about size, rather than “copy”. When this rule was set, 56 of 254 flagged coins had real markets of their own, so only the small ones are hidden. “Main” is judged among coins that are still trading, so the true original may be a dead coin that no longer appears. StonkFun's search matches plain text, so a copy spelled with look-alike letters is found from its own page but may be missing from the original's.

Read from Raydium LaunchLab's trade tape for the coin's pool, walked back to the first trade. First second: wallets that bought in the same second as the first trade; three or more is called a bundle. Snipers: wallets that bought one to four seconds after the first trade, leaving out the creator and the first-second wallets. Creator: what the creator wallet bought and sold on the curve, and the seconds from the first trade to its first sell. Percentages are of total supply, taken from Raydium's launch record or, failing that, the mint account; with neither, nothing is computed and the panel says so. “Still holds” is a wallet's buys minus its sells on the tape, never below zero. “Holds now”, when shown, is the wallet's balance on chain, read with the holder breakdown: it catches coins moved out as well as sold. Within one second, buyers are listed in the order the tape recorded them. An opening once read is kept: a later attempt that can no longer reach it does not replace it.

Where it goes wrong. The tape's clock is whole seconds, so “the first second” stands in for “the same slot”: on a hot launch, unrelated buyers can share a second. Only swaps are seen, so a wallet that moved its coins to another wallet still reads as holding, and a wallet that received coins without buying does not appear. Wallets are not linked to each other: a bundle of one person's wallets looks the same as several strangers. The tape ends at graduation, so for a graduated coin the figure is what a wallet held as the curve closed, and is worded that way. The tape only reads backwards from now, so a coin with more than about 3,000 trades on its curve is out of reach. Legacy launches have no such tape at all.

The creator is the wallet StonkFun's launch ledger records for the coin. Its launches, and how many were made in the last day, come from that ledger (the most recent 100). A launch counts as still trading when the coin is on Leverminal's board today, which means it has a pool worth the name or someone trading it. If the board cannot be loaded, what became of the launches is shown as not known, never as none. A wallet is described as launching by the batch at five or more launches in a day, or twenty in all. Across one ten-hour sample of 2,000 launches, wallets with five or more launches reached $1,000 of volume more often than one-time creators (59% against 47%) but a $30,000 peak half as often (1.0% against 2.0%).

Jupiter also counts how many coins a wallet has minted on every launchpad. Leverminal shows that figure, and lets you hide coins by it, only when Jupiter's wallet is the same wallet as StonkFun's creator.

Where it goes wrong. Jupiter reports the wallet that deployed a mint. For older StonkFun launches that is StonkFun's own deployer, which deployed thousands of unrelated coins; read naively it made 58 of 100 coins in one sample look like spam. Those launches are never judged by it. A wallet can also be a launch service used by many people, and one person can use a fresh wallet for every launch, which this cannot see. Many launches is a count, not an accusation.

Every five minutes Leverminal reads Jupiter's list of coins trending across Solana in the last hour. A StonkFun coin is marked when a coin with the same ticker key is on that list, is on a named launchpad other than StonkFun, and is worth at least $5,000. “Running on” means the other coin was created first; “copied on” means the StonkFun coin was. Unknown dates read as “running on”. If the list cannot be refreshed, marks from a list more than twenty minutes old are dropped rather than shown.

Where it goes wrong. It matches tickers, not ideas: two coins can share a word without sharing a story, and a narrative under a different ticker is missed. The trending coin may itself be a scam. Established tokens that Jupiter tags with no launchpad are ignored on purpose.

From Jupiter's token data: the holder count, the share of supply in the ten largest wallets with the pool left out, and net buyers, the wallets that bought more than they sold in the last hour. On the board these are looked up only for the coins that already passed your other filters, a hundred at a time, for at most 300 coins: past that, the first 300 in your sort order are judged (the 300 largest, when sorting by holders or buyers) and the board says so. A number is refreshed every ninety seconds when Jupiter's budget allows, and one up to fifteen minutes old is still used when it does not.

Where it goes wrong. One owner spread across many wallets passes a top-ten test. Holder counts include dust and airdropped wallets. Jupiter's request budget is shared with live trades, which always come first, so numbers can be missing or up to fifteen minutes old. A coin minutes old may not be known to Jupiter yet. Either way a coin without numbers cannot pass a holder or buyer filter, and the board says when the budget, rather than the coin's age, is why.

Every token account of the coin, read from the Solana chain in one request and added up by owner: a wallet with several accounts is one holder, and an empty account is none. From that: the holder count; the share of supply in the 10, 50 and 100 largest wallets; the creator's wallet as it is now; what sits in pools, the curve and other programs; and every wallet sorted into bands by what its holding is worth at the coin's price when the holders were read (by share of supply when there is no price). An address that no key can sign for, one off the ed25519 curve, is a program: a pool, a curve, a lock, a vault or a multisig. The 200 largest holders are checked; programs are listed in the table but left out of every figure, and so is StonkFun's tax wallet. The largest hundred wallets are also looked up on pump.fun, and a name is shown only when pump.fun marks the wallet as one of its own users; a name written to offend is hidden. The count is refreshed every ten minutes.

Where it goes wrong. One owner with many wallets is many holders, and a farm of fresh wallets looks like a crowd. The creator figure is one wallet: coins the creator moved elsewhere are not seen. A multisig holding a team's coins counts as a program, not a wallet. Worth is at one price, so a thin market overstates what a large bag would fetch. pump.fun names are what their owners typed, not an identity check, and pump.fun's API is undocumented: when it does not answer, names are missing, and the page says so rather than showing nobody as named. A big coin takes several seconds to count, and the public Solana RPC may turn the request away when busy; the last count is then shown with its age.

The whale page watches the 40 busiest coins on the board by 24-hour volume, from every quote asset, that have a real market behind them (a pool of $5,000 and $5,000 of volume, or $50,000 of volume where StonkFun gives no pool figure); the eight busiest coins still on their bonding curve with $5,000 of volume, which the first rule mostly misses because a busy curve rarely holds a $5,000 pool; and LEVERCAT whatever its volume. A graduated coin's trades come from GeckoTerminal: its top pool, and up to two more when the top pool carries under 80% of the coin's volume and the other pool has $2,000 of volume, 5% of the coin's and a $1,000 reserve. A coin still on its bonding curve is read from Raydium LaunchLab's trade tape instead, priced in dollars at the coin's price now. The wallet shown is the transaction's signer. Buy or sell is read from what the wallet gave and what it got: GeckoTerminal's own buy and sell label is about the pool's first token, which for LEVERCAT's main pool is xSOL.

A trade is whale-sized at the size picked on the page ($500, $1,000, $5,000 or $10,000; $1,000 unless changed), or at 1% of the coin's market cap and $250 or more: $400 into a $20,000 coin counts. Each trade says which reason applies. One wallet's legs of one coin in one transaction are one trade, so a swap routed through two pools is added up, and a transaction that buys and sells the same coin counts only what the wallet kept. Rotating into and out of are whale buying minus whale selling per coin over the window picked (one hour, six or a day), with the number of distinct wallets; a coin whose net is under a tenth of what whales traded in it is churn and is in neither list. A rotation is a wallet that sold whale size of one coin and then bought whale size of another within six hours, both inside the window; the amounts are the sells before and the buys after.

A wallet reads as a bot when, in one coin, it bought and then sold (or sold and then bought) within ten minutes three times or more, judged on every trade held from $250 up. Bots are left out of the trades, the flows and the rotations unless shown, and the page counts what it left out. Never counted at all: StonkFun's tax and reward wallet, which sells each reward coin's collected tax for the quote asset to pay holders and so reads as a steady seller of every one, and Leverminal's own treasury and fee wallet, whose LEVERCAT buybacks are not anyone's position.

GeckoTerminal documents about 30 requests a minute; from Leverminal's server it answered about five. The server learns the real figure and never goes past it: the busiest pool is read every minute and the others in turn, about every thirteen minutes at five a minute, and each read asks only for trades of $250 and up, so one read reaches back hours. The turn follows the money: every thirty seconds DexScreener says how much each pool traded in the last five minutes and the last hour, a pool that just traded $250 or more is read next, and one that has not traded $250 since it was last read waits, because nothing whale-sized can have happened there. A trade shows up a minute or two after it happens where money is moving, and later in a quiet pool; the page says how many pools it has read and how often. Nothing is stored beyond the server's memory, so a restarted server reads the pools again from the last day. For what the top holders of the largest reward coins are trading, The Stonk Board's Robot Mines follows them by wallet.

Where it goes wrong. A coin that is not watched, or whose pools have not been read yet, shows nothing, and that is never the same as no whales: the page lists the coins it watches. One wallet can be many people (a trading bot or service that signs for its users shows as one signer), and one person can be many wallets, splitting a buy under the threshold where it is not seen at all. A sell may be the end of a move to another wallet rather than an exit, since transfers are not trades. A rotation is two trades close in time, not proof that the money from one bought the other. The share of market cap uses the cap now, not at the time of the trade, and a curve trade is priced at the coin's price now. Only the watched pools are read, so a swap routed partly through another pool is undercounted (one MASK sale of $9,475 on 27 September read as $8,286), and trades under $250 in one pool are not read, so a swap split into small legs can be missed. The wallet is the one GeckoTerminal names as the trader; where a relayer pays a user's fees, that is the user, but a service that trades for many people is one wallet here. The bot rule catches wallets that flip quickly; a slow wash trader, or a person who scalps, is judged the same way as everyone else.

Read directly from the coin's mint account: whether more supply can be minted, whether accounts can be frozen, and any Token-2022 extension that gives someone else power over a holder's coins (a permanent delegate, a transfer hook, accounts frozen by default, non-transferable). For a taxed coin: the rate the chain charges, the wallet that can change it, and the wallet that collects it. On a StonkFun reward coin both are StonkFun's platform wallet, which pays the tax out to holders; that is every reward coin by design and is not flagged. A tax controlled by any other wallet is flagged, and so is one nobody can collect. When the mint account cannot be read, none of this is claimed: the card gives StonkFun's record of the rate and says who controls it is unknown, and shows no all-clear.

Where it goes wrong. If StonkFun ever rotates its platform wallet, every taxed coin would be flagged until this site is updated.

Every coin launched on StonkFun through Raydium LaunchLab is minted with exactly 1,000,000,000 tokens at six decimals, and its mint authority is revoked, so the supply can only go down, and only by a burn. The burned share is what the mint account's supply is missing from that billion, as a share of it, counted to the last base unit. It counts every burn, whoever made it. On 27 September, 104 of 190 LaunchLab coins still had the full billion and none had more.

StonkFun's own figure, shown beside it as Flywheel burns, is its ledger of the burns its flywheel made, and leaves out every other burn. The two differ, and both are true: on 27 September LEVERCAT's ledger counted 4,460,738 tokens in 1,996 burns, while 6,296,604 were missing from its billion.

Where it goes wrong. Legacy launches ("Raydium" rather than "Raydium LaunchLab") are nine-decimal coins with supplies from 815 to 990 billion and no record of what they started with, so no share is shown for them rather than a guess; their flywheel ledger still is. A LaunchLab mint that breaks the pattern, with more than the billion, other decimals or a live mint authority, shows nothing either. Tokens sent to a dead wallet or locked in a contract are not burned and are not counted. Locked supply is not shown at all: no keyless source measures it. The one lock Leverminal can read, LaunchLab's own vesting for the creator, is shown with the opening.

StonkFun's list of assets a coin can be launched against carries no dates. Leverminal relies on two things instead. The list is in the order pairs were added, newest last: along its last sixteen entries, the age of each pair's first launch fell without exception, from 3.7 days to two hours. And people launch against a new pair within minutes of it appearing, so the first coin launched against a pair dates the pair. A pair is shown as new when nobody has launched against it yet, or when the first launch against it was in the last three days. The list is read every minute, from its end backwards until pairs stop being new. "Spotted" is shown only for a pair that appeared while this server was running.

The counts come from two of StonkFun's records, not one. Their token index is what the rest of the terminal reads, and it has been seen to hold none of a new pair's coins: on 20 September about two hundred coins were launched against FTT within minutes of the pair appearing, and the index answered that the pair had none. So every minute Leverminal also sweeps the last few hundred rows of StonkFun's launch ledger, which had all of them, and counts them by what they were launched against. Whichever record saw more wins. A count off the ledger is written ≥201, because the sweep covers a window and not all of time.

Where it goes wrong. "Live for" is the age of the first launch, not of the pair: a pair nobody launched against for a day would read a day younger than it is, and one that sat unused for longer than three days, then got its first coin, would read as new. The order of the list is observed, not promised by StonkFun; three old pairs in a row end the search, so new pairs listed out of order deep in the list would be missed. A pair with many coins already is not an open field, however new: the counts are beside it. When the ledger is the only record of a pair's coins and the sweep does not reach back past the first of them, the pair reads "live, age unknown" rather than being dated from the oldest row the sweep happened to see.

Final stretch is StonkFun's own “about to graduate” flag, or 80% or more of the curve; the filter, the chip and the export all use that one rule. Just graduated is graduated within the last day. The curve-progress filter and sort apply to coins on a curve: a graduated coin is off the curve, not at the top of it. A market-cap band includes its floor and excludes its ceiling, so neighbouring bands never share a coin, and "Under $250k" means below $250,000. Volume ÷ market cap is the day's volume over the coin's market cap, the number the Turnover sort ranks by: "20%+" is a coin that traded at least a fifth of its market cap in 24 hours. Dust, hidden unless asked for, is a graduated coin with under $500 of pool liquidity, or a curve coin that has not moved past its starting mark and has under $100 of daily volume. StonkFun reports no liquidity figure for older launches; those are judged by whether they trade.

Where it goes wrong. Volume is StonkFun's 24-hour figure and counts every trade, including a wallet trading with itself, so a high ratio can be one bot. On a coin worth $3,000, a single $1,500 trade is 50%.

A coin is a cat when a cat word is in its name or ticker: cat, kitty, kitten, meow, neko, purr, nyan at the start of a word, kat at the end of one (NEARKAT), and the Japanese and Chinese words for it. Names are folded the way tickers are above, so look-alike letters do not hide one. "Cat" inside a longer word counts when it ends the word or runs into a consonant (LEVERCAT, STONKCATS, catwifhat), and not when English carries on after it (catalyst, education, scatter, catch). Checked against StonkFun on 27 September: of 10,713 coins with "cat" in the name or ticker, 10,128 are called cats.

Where it goes wrong. It reads words, not pictures: a cat called Whiskers is missed, and a coin can say cat without being one. A few cat puns that look like English (CATE, CATOSHI, CATINU) are let through by name, so the next one will be missed until it is added. Meerkats count, because NEARKAT does.

Leverminal does not model yield, APR or what holding a reward coin pays. That is the work of The Stonk Board, a friend of Leverminal in the same ecosystem: it ranks the top 100 reward coins bonded a week or more and explains how it models their APR. Its Robot Mines shows what the top holders of the top coins are trading. Token, holder and creator pages link to a coin's page there whenever its list includes the coin, read from the sitemap it publishes, so no link lands on a missing page. Leverminal covers a coin from launch through its first week, and the coins outside their list.