We ran 7 crypto strategies with real money. All 7 lost.
Not a paper exercise. We funded a live wallet, wired the execution, and let the machines trade. Here is every strategy, every result, and the one that came closest to working.
The seven
| Strategy | How tested | Result |
|---|---|---|
| Memecoin sniping (new launches) | Live capital, ~15 snipes | Every position faded |
| Micro-scalping large caps | Live + 60d backtest | Spread and fees ate every edge |
| Momentum — large caps | 64d, 12 tokens, hourly | −12.2% |
| Momentum — top 24h gainers | 60d, 60 tokens, hourly | −95.7% |
| Grid trading (range harvest) | 67d, SOL/USDC, hourly | −4.2% (0 round trips in final week) |
| Trend-following (10d MA switch) | Live, armed, real rotations | Protects crashes, bleeds in chop |
| Mean reversion (buy oversold) | 62d, 7 tokens, hourly | −9% to −22% |
The one that hurt most to publish
Chasing the top 24-hour gainers — the strategy that feels most obviously correct when you look at a list of tokens up 30% — returned −95.7% over 60 days across the 60 most-liquid Solana tokens. It won 26 of 94 trades.
The reason is visible in the data the moment you look past the headline number: the token up 31% today is frequently down 31% on the week. A gainers list is not a list of things going up. It is a list of things that already went up, photographed at the top.
The most useful number we found
This is the trap behind every “90% accurate” bot you have ever been sold. Win rate is the number that markets well and means the least.
What actually did work
Two things, both unglamorous. Trend-following didn't make money, but it cut losses — the same result we found across 30 years of stock data. And fee income(providing liquidity, lending) was the only positive-expectancy activity we measured, because it gets paid by the same choppy market that punishes everyone trying to predict it.
The honest conclusion is uncomfortable for anyone selling trading products: in a market that isn't trending, the absence of edge isn't a strategy problem you can solve by finding a better strategy. Nobody is being paid. The correct move is to stop paying fees to find out.
Why publish losses?
Because everyone publishes wins, which is why nothing you read about trading is useful. We would rather be the people whose numbers you can trust when we eventually say something worked. Every test above is reproducible from public data, and our live positioning is published in real time.
We test one strategy people swear by and publish whatever the data says — including when it says the strategy doesn't work. Free, no pitch.