What "graduation" actually measures
A pump.fun token launches onto a bonding curve. Buys push the price up the curve; when roughly $69,000 of market cap has accumulated, the curve fills, liquidity migrates to a real AMM pool, and the token "graduates". Graduation is the milestone the whole ecosystem treats as the dividing line between a joke and a real token.
It is a throughput measurement. It says a certain amount of money went in. It says nothing whatsoever about whether that money can come back out, or about who was holding when it did.
The distribution
Across the 32,260 launches where we could observe a peak:
| Percentile of launches | Peak market cap reached |
|---|---|
| Median (50th) | $4,000 |
| 90th | $16,000 |
| 99th | $195,000 |
The 99th percentile launch peaks at $195k. To put that in perspective: a launch has to finish in the top 1% of all launches just to reach three times the graduation threshold. The median launch never gets past $4,000, which on a 1B supply is a price of $0.000004.
The finding: graduating is not surviving
817 launches reached ~$69k at some point. Here is where they are now.
| Outcome | Launches | Share of graduates |
|---|---|---|
| Reached ~$69k, now trades under $5k | 673 | 82.4% |
| Reached ~$69k, still at or above $5k | 144 | 17.6% |
Four out of five tokens that cleared the bar the market uses to separate real from fake are now worth less than a used car. And the collapse is not confined to marginal graduates that scraped over $69k and rolled back:
| Peak reached before collapse | Launches now under $5k |
|---|---|
| Over $1,000,000 | 68 |
| Over $5,000,000 | 32 |
| Over $10,000,000 | 16 |
Sixteen launches in this sample printed an eight-figure market cap and now trade under $5,000. That is a drawdown of more than 99.9%, and every one of those charts looked like a winner while it was happening.
This is the case for deployer history in one sentence. Every check that asks "is this token real?" — liquidity, holder count, market cap, graduation status — is answered yes by a token at the top of that move. The only thing that separates the 68 from a genuine run is who launched it and what they did last time, and that is knowable before you buy.
What a collapse looks like up close
We re-derived the candles for the largest collapses individually rather than trusting a stored number. Two representative shapes:
# A sustained run that still ended at zero peak $130.0m 2nd / 3rd best hour $121.3m / $109.7m hours >= 50% of peak 7 # a real market, not one print median hourly close $46.2m total candle volume $1.2m now $2k # A four-hour life peak $72.2m 2nd / 3rd best hour $61.1m / $1.4k # third hour is already dead hours >= 50% of peak 2 median hourly close $1.4k total candle volume $701.5k now $0
The second one is the whole life cycle of a token in four hourly candles: two hours at a nominal $60–72m, then nothing. Roughly $700k of real volume traded through it.
A necessary caveat on the word "market cap". These figures are last price multiplied by the 1B token supply, which is how pump.fun, DexScreener and every aggregator quote them. It is notional. Nobody extracted $130m from that token — the total volume that traded through it was $1.2m. On a thin pool, a modest amount of buying moves the quoted market cap enormously.
We quote it because it is the number a buyer sees on their screen at the moment they decide. The gap between "the screen says $130m" and "$1.2m has ever changed hands" is the risk.
How concentrated is this in repeat launchers?
| Wallets in sample | Count | Share |
|---|---|---|
| Have 2 or more launches that reached ~$69k | 147 | 5.2% |
| Have never had a launch reach ~$69k | 2,288 | 80.2% |
147 wallets have done it more than once. That is the population worth knowing by address, and it is the population our serial-launcher index exists to publish. Note what this does not say: repeat graduation is not proof of wrongdoing. Some of those wallets are simply prolific. It is a fact about a counterparty, not a verdict on one.
Method
- Corpus. 14,597 deployer wallets observed by Cabal-Hunter between 14 June and 14 August 2026. Wallets enter the index by being scanned, so this is a sample of wallets people asked about, not a random sample of the chain.
- Peak market cap. pump.fun hourly USD candles over the token's full trading life, taking the maximum hourly high multiplied by the standard 1,000,000,000 supply.
- Current market cap. DexScreener, taking the pair with the deepest liquidity.
- Graduation threshold. Taken as $69,000. The real figure drifts with the SOL price; we use a fixed constant, so launches within a few percent of the line could fall on either side.
- Sample. 32,260 launches from 2,852 wallets, measured 14 August 2026.
What we excluded, and why it matters
Our indexer fetches candles for at most 20 launches per wallet and inside an 8-second budget. When that budget is missed it falls back to using the current market cap as the peak. For a dead token that produces a peak of roughly zero — which, if we counted it, would appear in the data as "never came close to graduation" when the truth is that we never looked.
| Launches in the index | Count | Share |
|---|---|---|
| Peak genuinely observed from candles | 32,268 | 81.9% |
| Fallback fired, peak equals current — excluded | 2,425 | 6.2% |
| No data at all — excluded | 4,710 | 12.0% |
| Total excluded as unmeasured | 18.1% | |
A separate audit then dropped 11 mints that reported large market-cap figures with no on-chain pair behind them; 8 of those fell in the observed bucket, leaving the 32,260 launches this article is based on.
Counting the 7,135 unmeasured launches as failures would have pushed the headline failure rate up, which is the direction that flatters this article. We excluded them instead.
One residual bias, stated rather than hidden: a token sitting at its all-time high right now legitimately has peak equal to current, and our exclusion rule drops it. That discards live winners, so the 97.5% figure is if anything conservative — the true survival rate is marginally better than we report.
What this does not show
- Not a random sample of pump.fun. These wallets were scanned because somebody was interested in them, which plausibly over-samples suspicious ones. Treat the rates as descriptive of this corpus, not as a chain-wide census.
- No intent is established anywhere in this data. A token collapsing is not evidence that its deployer rugged it. Most launches fail because nobody wanted them.
- Snapshot, not a rate. "Now trades under $5k" is measured on 14 August 2026. Recent launches have had less time to fail than older ones.
- Per-wallet sampling. With a 20-launch cap per wallet, the most prolific launchers are under-sampled relative to their true output.
The practical takeaway
If you are trading these, or building an agent that does: stop using graduation as a safety signal. It is a liquidity milestone that 2.5% of launches reach and that 82.4% of reachers do not survive. Holder counts, market cap and liquidity all look their best at precisely the moment they are least informative.
The one input that is available before the move, and that does not improve as the chart improves, is the deployer's record. That is the entire reason we index it.