Crypto trading research
We test trading ideas. Most of them fail.
Construct Labs is an independent research lab. We test popular crypto trading ideas on free public market data, with real fees and slippage, and judge each one only on data it was never tuned on. Every result gets published, including the ideas that lost money.
Questions or study ideas: hello@constructlabs.io
Results
| Idea | Test result |
|---|---|
| Buy small coins within 1 min of a listing post | +10.4% /trade |
| Best of the 10 strategies picked on older data. Those 10 averaged +1.8%. | |
| Short new listings for 90 days, BTC hedged | +26% /yr |
| Crowded: shorts opened in 2026 paid 22% in funding on average. | |
| Altcoin breakouts, best training setting | −29% /yr |
| Bitcoin buy-and-hold made +55% a year. | |
| Pump.fun tokens that bounce after a dump | −5.4% /trade |
| Buy when exchange wallets move a token early | −0.5% /trade |
| Bots sell at round multiples | no pattern |
| Made money after costs | 2 of 6 |
How we test
The rules every study follows.
- 01
Judged on unseen data
We build on older data and score on newer data the idea never saw. Only the second number counts.
- 02
Costs in
Exchange fees plus slippage that grows as liquidity shrinks. A thin coin costs more to trade than bitcoin, so it does here too.
- 03
Delisted coins stay in
Dropping coins that were later delisted makes almost any strategy look better than it was. We keep them in.
- 04
No hindsight
A signal can only use data that existed at that moment. If a backtest needs the next candle to work, it's thrown out.
- 05
Enough trades to count
Fewer than 20 trades in the test period and a result doesn't count, however good it looks.
- 06
Free data only
Public exchange and blockchain data, so anyone can check our work.
Studies
Six so far. Each has its numbers, method and weak spots on the research page.
Listing announcements
Buying within a minute of a listing post paid on small coins. Waiting an hour turned it into a loss.
Held upNew listings vs bitcoin
87% of new Binance listings since 2023 trailed bitcoin over their first 90 days. Shorting them worked on paper, but the trade is crowded and getting expensive.
Held upAltcoin breakouts
Strong from 2018 to 2022, losing from 2023 on. None of 288 settings beat holding bitcoin.
FailedPump.fun second legs
About 1 in 6 dumped tokens doubles again. Not often enough to cover 5% trading costs each way.
FailedOn-chain listing tells
Some exchanges do move a token into their wallets hours or days before the post. Buying on it still lost money.
FailedBot exits
We looked for pump.fun sellers clustering at 2x and 3x. The pattern isn't there.
No pattern
Settlement tracker
Our first tool. Now in testing.
Many hourly and 15-minute crypto event contracts don't settle on the price in your exchange app. They settle on the average of 60 one-second readings of an index built from several exchanges' order books. That's how a contract can look safe at 59 seconds and still flip.
We rebuilt that index from the same public order books and run it live. The tracker shows it second by second, how much of the final average is already locked in, and what the rest of the minute would need to do to change the result.
How it works →Get the tracker when it opens.
Bitcoin first, then ETH, SOL and XRP. One email when it's live. No newsletter.