The system with the best pedigree
Buy when price closes above the highest high of the last N bars, exit when it closes below the recent low. That is the Donchian channel breakout, and it is the skeleton of the Turtle experiment — the most cited proof that a mechanical system can be taught to novices and still make money.
That pedigree is exactly why it deserves a real test rather than a chart screenshot.
What we tested
Our full backtest window on the top 50 coins by market cap, with fees and slippage applied to every trade (the exact window is shown on the settings page). Then the exit grid: 5 stop-loss levels x 6 take-profit levels = 30 full backtests. The live grid refreshes with our data pipeline, so the exact cells move; the shape does not.
The full grid is live here: Donchian Breakout — stop-loss & take-profit settings.
Result: 0 of 30 combinations profitable
Every cell in the grid came out negative. As of 2026-07-26, the highest profit factor anywhere in the surface was 0.80. A profit factor under 1.00 means the losing trades took more than the winning trades brought in. It is the number to look at first, because unlike a return figure it does not depend on how you sized positions.
Trade counts ran from about 6,000 to 11,500 across the 50-coin universe, so this is not a small-sample accident. There were enough breakouts to be sure the strategy had its chance.
The 59% win rate that still lost money
The single highest win rate in the grid — 59.0% — belongs to a wide 12% stop with a tight 4% take-profit. It wins nearly six trades in ten and still loses; it ranks 24th of 30 cells by return, so it is in the better quarter of a table where every cell is negative.
The obvious explanation is the win-rate trap: bank 4%, risk 12%, and one loss erases three wins. That is the standard story, and for this cell the data does not support it.
Here is the actual exit breakdown for that cell: of 8,446 trades, 53.5% closed at take-profit, only 11.3% were stopped out, and 35.3% timed out at the 48-bar limit. Multiply it through and the take-profit column (4,516 × 4) is larger than the stop-loss column (952 × 12). The stops are not what is eating this strategy.
Where the money actually goes
If wins outnumber losses and stops rarely trigger, the loss has to come from somewhere else. Two places:
The timeouts. More than a third of trades never reach either exit and close at whatever price exists at bar 48. Those are not clean small losses — they are unmanaged outcomes, and they are the largest uncontrolled block in the table.
The turnover. This cell alone settles 8,446 round trips. Fees and slippage apply to every one of them. A strategy can be right more often than it is wrong and still hand its edge to the exchange if it trades that much.
That is a more useful lesson than the win-rate trap, because it is less visible: the trap at least shows up in your average win and average loss. Timeout drag and turnover cost do not appear in either number.
Why breakouts specifically struggle here
A breakout system is a bet that the move continues. It pays for that bet with many small false starts, and it needs the winners to run far enough to cover them.
The grid says the runners never paid. What is more telling is how they failed to pay: the surface is flat. Averaging profit factor by stop level across the 5% to 12% stops shows no ordering at all — no gradient, no direction, no setting that helps. (Exact values move with each data refresh; the flatness is what has held.)
That matters, because a flat failing surface is a different diagnosis from a sloped one. Our mean reversion grid improves monotonically as the stop widens, which tells you what is hurting it, and the Heikin-Ashi grid is flat in the same way this one is. Donchian shows nothing to tune: widening the stop to give the trend room does not help, and tightening it to cut the false starts does not help either. The premise itself is what did not survive this window.
”You used the wrong settings”
That objection is the reason we sweep grids instead of publishing one configuration. Every stop-loss and take-profit pair our engine supports is in the table, and the whole surface is negative. A strategy that only works in the one cell nobody has tested yet is not a strategy.
This is consistent with our broader research: simple directional signals on crypto OHLCV have not survived honest out-of-sample testing with costs.
The honest takeaway
The Turtle pedigree is real, and it came from markets and a risk framework that are not this. As a standalone crypto strategy with fixed exits, Donchian breakout did not produce a profitable configuration in this window.
Check the live numbers yourself on the settings grid, or build your own version in the free simulator and see what you get.