How it works

Every figure on a result page comes from a swap that happened. Here is the path from the chain to the table.

The range picks wallets, not numbers

The range decides which wallets appear. Their numbers come from all of that wallet's trades on the token: before the range, inside it, after it.

So a wallet can show a bigger total than what it spent in your range: it was buying earlier too.

The scope switch above the table recounts the same stored trades up to 24 or 48 hours after the range. It costs nothing: the trades are already here.

Fresh pumps on the home page

Tokens created in the last 24 hours with a market cap of $100K or more, $20K of liquidity, $50K traded in a day and 100+ holders. Clones with huge caps and no trading, and charts bought by one bot, are filtered out. Each one shows its market cap now and the peak it reached. Biggest peak first; the list is refreshed at most every 10 minutes. A click opens the chart, where you mark the range.

Ready lists

Top traders on the home page: up to ten wallets that made the most last month and trade like people, for when you have no token to start from. The September list stays the same all October; on the 1st of the month (UTC) the next one is counted. Beside it, a card says why it is worth following.

The candidates are known KOLs and the best of the market from Solana Tracker's rankings. How much each one made, tracced counts itself, from the wallet's own swaps over the whole month, the same way as its card. The rankings count as profit the tokens that came from another wallet with no purchase; we leave out a sale with no purchase found under it. A swap of a memecoin into a tokenized stock (RACE, SKHY, SPCX) is a real exit and counts at the stock's price.

Only people stay, not machines. A wallet is left out when, in the month, it made more than 1,500 swaps or traded more than 300 tokens, closed fewer than 5 positions, won less than 30 % or more than 90 % of them, held a position less than 2 minutes on average or closed most of them within a minute (a sniper's pace), started trading less than 30 days before the month began (a fresh wallet), or is marked a bot, an exchange or an arbitrageur. Of the rest, the list keeps the ones that made more than $10,000 in the month, best first.

The list's page opens with a connected wallet. Its rows are the month's numbers; a row opens the wallet's card, the same card as on a result, with its last 7 and 30 days, and under them Best shows the tokens it made its profit on. Beside the list, Best ROI ranks the same kind of people by the profit on what they bought in the month (at least $2,000 in and $5,000 made), with its own candidates from Solana Tracker's ranking by return. Follow puts a copy of the list in your watchlist, under the same name, with the bells on while your account has bells left. The copy is yours: next month's list does not change it; your watchlist says when the new one is out.

The numbers under the chart

One strip under the chart, a number and a word each. A click on a number shows exactly those wallets, or sorts by it.

Always there:

• bought: the wallets that bought in the range, the list below. Its i says how they were counted.
• spent: what they spent in the range.
• sold out: the share that sold 99% or more of what they bought.
• holding: the share that sold less than half.
• best ROI: the highest average exit cap ÷ average entry cap among them.

Then up to four findings, each only when it says something:

• creator bought: the token's creator bought in the range, and how much.
• bundled: wallets that share a funder made that share of the range's buying (shown from 10%).
• fresh: of the wallets checked for age, the share under a day old when they bought (shown from 10 wallets and 10%).
• to top 10: the ten most profitable wallets' share of all the profit made (shown from 20 wallets in profit and 40%).
• seen before: with a wallet connected, wallets that were early in your other saved analyses too (shown from 3).
• from exchanges: of the wallets whose first funder is known, the share funded straight from an exchange (shown from 10 wallets and 5%).
• in profit: wallets with a realized profit (shown from 20 wallets).

Where the numbers come from

Every row is a swap recorded on chain. Market cap is the swap price times the token supply, so an entry cap is the cap at that exact trade.

The chart can have a gap, often right after a token leaves pump.fun, and that is usually the pump itself. Once a range is analyzed, the chart fills the gap with candles built from that analysis's own trades.

Held is how long a wallet held what it sold, on average, weighted by amount. A wallet that sells $999 of its $1,000 two seconds after buying and keeps the last $1 for a day held about a minute and a half, not two seconds and not a day. What it still holds is not counted: the table shows it apart. The card's average hold over 7 or 30 days is counted the same way, position by position.

The range is always read in full. The i after the counts above the chart says for how many wallets exits are known, and how many traded in the range but stay out of the table (too small, or only sold); it turns into ! when exits are not known for every wallet.

How long it takes

A busy token takes seconds, not minutes.

The result page stays light however many wallets it holds. The table arrives as numbers, the first hundred rows are drawn and more on request, and sorting, filters and export work on the numbers, so a result with thousands of wallets opens and sorts at once, on a phone too.

Dollars or SOL

The USD | SOL switch above the table changes the unit of every amount that came from a swap: spent, sold, made.

Nothing is converted at today's rate. Both figures were recorded by the same swap, so the SOL amount is what actually moved on chain and the dollar amount is what it was worth then. Profit is counted the same way in both units, from the cost of the tokens actually sold, so the two never disagree.

What a wallet still holds stays in dollars. It is a valuation at a later price, not a swap that happened.

What the token itself did

Two events are marked on the price itself: M where trading left the launchpad, the DexScreener logo where someone paid DexScreener to show the token's profile. Each badge sits above the candle where the event happened; hover it for the time and the detail. ⇤ Launch under the chart brings the first hours of trading into view. Together the two events often explain the timing of a run.

Age and funding

Two facts are not trades: a wallet's age is its first transaction ever, and its funder is whoever sent it its first SOL. They are what fresh and bundle are built from. The same check reads the transaction right before the wallet's first buy here: when it is a week old or older, the wallet slept until this token and gets dormant.

The first 200 wallets by PnL, the ones that made money on the token, are checked in the background after the table is already on screen. Any other wallet is checked the moment its card is opened, and the answer stays in the result for everyone. When a funder cannot be found, it stays empty instead of guessed.

A bundle is three or more wallets here whose first SOL came from the same wallet. An exchange or an app funds thousands of wallets that have nothing to do with each other, at any time. From such a busy funder, only the wallets it funded close together in time form a bundle: that is how one operator creates wallets for a launch. The rest stay out of any bundle. In the PAID demo most of what looked like bundles was one app's service wallet (154 of the buyers) and an exchange (33). On another token, one wallet created 200 wallets in 41 minutes, and all of them bought within minutes of the launch: that is still a bundle, however busy its funder.

When the first SOL came straight from a known exchange, the wallet's card says "funded by" and the exchange's name instead of an address. From an exchange, as from any busy funder, only wallets created close together make a bundle. Funders our own list of exchange wallets does not know are looked up in the labels of InsightX: the card then says the funder is an exchange, an app or a casino, such a funder makes no bundle, and its wallets count among those funded from exchanges in the strip under the chart. The names stay with InsightX.

In the table a bundle's tag takes one colour per funder, so two bundles side by side are told apart; a click on the tag shows that bundle alone.

When nearly all of the top 200 by PnL are fresh or in bundles, one hand ran the launch and little in the list is organic buying. A red word, first among the findings under the chart, says which kind:

Until the check reaches a wallet, fresh, bundle and the filters that hide them do not know about it yet. Below the first 200 they stay unknown unless a card is opened: a bundle there counts only the wallets that were checked. When the check is paused, the page says so.

Organised launches

After the table, tracced checks how the token was launched, from two free sources, and adds only facts with a proof (owner, 14.10). There is no score.

• Jito bundles of the first 3 minutes. Each early sale is looked up at Jito, first those with another wallet's buy in the same second: when that wallet bought in the same bundle, both get own-bundle. Wallets in the bundle of the transaction that created the token get launch-bundle.
• The creator fee. pump.fun lets a creator send the fee from every trade to another wallet. tracced reads where it was sent and how much SOL it has paid out so far.

When any of it is found, the strip above the table says organised launch, with the facts on hover; a click shows exactly those wallets. Such wallets are never in the ready lists of top traders.

On one launch we checked by hand, 14 of the 20 most profitable early wallets belonged to the launch itself. These checks catch the ones that traded inside the launch's own bundles; wallets tied to it only by later shared signatures are not caught yet.

The wallet card

Click a wallet's address, or anywhere else in its row. The chart mark before it, which comes out with the star when the pointer is over the row, puts its buys and sells on the chart instead, in the wallet's own colour. In the card, a click on the address copies it. On a phone each wallet is one line, its PnL and ROI on the right; a tap opens the card from the bottom of the screen, and a tap above it or pulling it down closes it. The card shows, top to bottom:

1. who it is, when the wallet is publicly identified (a star for a KOL, its X account, the app it trades through), and next to the address its age, like 94d old, and who sent it its first SOL (funded by …); hover either for the detail. A wallet outside the first 200 shows … for a moment while its card checks it; when that cannot happen (no wallet connected, the day's card checks used up) it shows ? or 1k+ tx: its age stays unknown rather than guessed;
2. its tags, ours and your own;
3. how it trades on every token over the last 7 or 30 days: realized PnL and its curve, win rate with wins and losses, volume, buys and sells, how many tokens and how many still open, average hold, best day, drawdown, how its closed positions ended and the hours it trades. Point at the PnL curve for any day's result and the running total after it;
4. the analyses you saved where it was early too, if any;
5. its trades, with a switch: on this token, its numbers here (bought, sold, PnL, ROI, held) and every trade; on all tokens, the latest tokens it traded. Click one of them: its chart opens right there, on the hours of the wallet's trades, with its buys and sells marked. The trades come with the 30 days; the candles cost what any chart costs.

Bought, in the table and in the card, is everything the wallet bought of this token: before the range, inside it and after it. Under it: what it bought inside the range, and the market cap there. Sold, PnL and ROI count the same buys, so the four numbers add up: PnL is what the wallet received minus what the sold tokens cost at its average price.

Filters you use often can be kept: open Filters, press "Save these filters" and name them. The set (the numbers, the hidden tags and the order) stays in your browser and applies to any analysis in one click.

ROI, in the card and in its own column of the table, is the average exit cap over the average entry cap, buys before and after the range included: 2× is +100%, 61.5× is +6,050%. Hover it for the percent.

The 30 days are counted by tracced from the wallet's own swaps, the same way as the table: average cost, and profit only on what the wallet actually bought. Profit counts on the day of the sale, as a broker counts it: a token sold in those days counts in full, at what the wallet really paid for it, even when the buy came earlier. For such a token the card reads the wallet's earlier trades of it too, up to 25 tokens with the biggest sales. A position is closed once 99% of it is sold; win rate is the positions closed in those days that made money, out of all of them.

A token sold in those days with no buy found is left out and counted apart: it came by transfer, or its buy is older than what the card reads, so its cost is unknown, and a transfer is not a profit.

A card someone already opened today opens free for everyone. The 7-day view is counted from the same swaps.

The AI agent

The agent on a result page reads that analysis and nothing else. Open it and pick a question (who sold the top, whether it was a bundled launch, who is still holding) or ask your own, in any language. Summary of this pump writes three cards: what happened in the range, what a buyer should weigh, and whose wallets stand out by the method written under the card. A wallet it names opens its card.

It reads only what the site computed from the table and does not compute a number of its own. Before an answer reaches the page, every number and wallet in it is checked against the analysis, and a line that fails is dropped. It never tells you to buy or sell, never predicts a price and never calls a wallet good or a token safe.

Ask who a wallet is, or who owns it, and the first line comes from tracced, not from the model: its X account or that it has none, a KOL or an exchange when the labels say so, a bot-like trader when its trades here look like one, otherwise a trader, then the app it trades through and the exchange its first SOL came from.

It knows your watchlist too: when you ask, it sees which of your saved wallets bought in this range, the lists that hold them and your own tags for them, so "which of my wallets are here?" or "what did my insiders do?" has an answer. When some of them did buy here, the panel offers that question as a button. It sees only your own watchlist, and only when you ask.

It answers only about the analysis it is on. Anything else, including a request to change how it works, gets one fixed reply. The agent needs a connected wallet; cards someone has already opened are kept, so opening them again is free for everyone. Questions go to AI model hosts whose policy is not to collect them or train on them; when they are unavailable, to Claude, Anthropic's model, which does not train on them and keeps them only for a limited time under its terms. Do not type anything personal into it.