Capo’s research lab · weekly

The Lab Drop

What our paper tests on real pump.fun data actually showed, including everything that lost money. Rules frozen before testing, judged after costs. No signals, no calls.

Research, not advice. Past results don’t predict future ones. NFA · DYOR.

Drop #1

what happens after a pump.fun coin hits $100k

We followed pump.fun graduates that crossed a $100k market cap. About 1 in 5 were still at $100k a day later. Four ways we tried to trade it all lost money after costs.

What we found

20.7%

still at $100k+ a day later. Of 353 pump.fun graduates we could follow for 24 hours after they first crossed a $100k market cap, 73 (20.7%) were still at or above $100k. Only 50 (14.2%) never dipped below it in our snapshots.

−97.9%

median move in 24 hours. From the $100k cross to 24 hours later, the median coin's market cap fell 97.9%. 78.5% were down by half or more. 11.3% were up at all. (Market-cap change only, no trading costs.)

23 min vs 1.4

the ones that lasted crossed later. Coins still at $100k a day later took a median 23 minutes after graduating to cross it. The ones that died crossed in a median 1.4 minutes. Their first hour also looked different: $41k vs $16k volume, and top-10 wallets holding 48% vs 98% at the start.

58%

of $100k crosses came within 5 seconds of migration. In a larger set of 10,282 graduates, 5,091 (49.5%) crossed $100k, and 58% of those crosses happened less than 5 seconds after the coin migrated. Still at $100k+ 10, 20 and 30 minutes later: 56.1%, 44.1%, 36.6%.

What didn’t work

  • Buy the $100k cross and hold0 of 1,440 rule sets made money after costs. The one we locked in before testing lost 10.1% per trade at 0.02 SOL size (278 trades), and 7.4% at 0.50 SOL. It beat random entries and still lost.
  • Filter for “stayers”Using the stayer traits above as entry filters: 0 of 80 and 0 of 55 rule sets positive. On newer data every filter did worse than no filter. Late crossers held up more often in-sample (33.6% vs 20.7%), but that didn’t turn into a trade.
  • Fast entries right after graduation448 entry and exit combinations, none with a positive median across test periods. Just getting in and out cost 6.9% round trip at 0.02 SOL and 4.8% at 0.50 SOL. Buying dips was the worst entry type (median −21%).
  • Our own asymmetric hold rule (V5)As frozen it lost 20.8% per trade over 916 trades, worse than random entries. None of 3,265 tuned variants went positive; the best still lost 5.2%.

So what

  • A $100k print is not a hold signal. Most coins gave it back within a day.
  • Costs are the silent killer: about 4.8–6.9% round trip before you’re right or wrong.
  • What looked like a pattern in hindsight (late, organic crossers) didn’t survive out-of-sample testing.
  • Because none of this made a tradeable edge, we turned it into checks you run before you touch a coin: see the Red-Flag Checker.
How we tested

Paper research only, no live trades. Rules were written down and frozen before each test, run on training data only, and judged after realistic costs, against random-entry baselines and walk-forward splits. Holdout data stays untouched. Across our research registry we have now tested 249,566 configurations; none has been promoted to live trading. 24-hour figures use July snapshots; minute-level figures use pool trades up to 45 minutes after graduation. 4,424 coins with corrupt price prints were removed. Past results don’t predict future ones.

Data provided by [Slink21]. Pumpfun Solana Dataset (2026). (Slinky21/Pumpfun_v2_dataset, CC-BY-4.0.)