The strategies, and what killed the rest.
We publish the research, not just the winners. Each strategy below shows what it does, how it was tested, and where it fails. Most of what we test never makes it here.
Straddle — Donchian Breakout Shipping
Direction-agnostic. Brackets the 20-day channel with a buy-stop above and a sell-stop below, cancels the loser on fill, and trails the winner with ATR. You never pick a side.
- USD/JPY H1 and H4; gold H4
- Both data halves positive
- Whipsaw taxed a full −1R in testing
Where it fails: ranging markets. Tested −EV on EUR/USD, GBP/USD and USD/CHF — those pairs chop and the brackets bleed. Backtested only; not yet live-verified.
Donchian Regime Gate Shipping
The single-direction breakout, gated by an ADX regime filter so it stands down in chop — its known weakness. Built after the Strategy Lab killed everything else on gold.
- Stop set inside the entry order
- ATR trailing, daily loss cap
- Sensitivity-tested, not curve-fit
Where it fails: extended ranges, and any period without a trend. Textbook ADX 25 over-filtered and cut returns — we use 15 because the sweep said so.
Mean-Reversion Scalper Retired
Faded Bollinger and RSI extremes on M5. It felt right and it lost money — across every pair we tried. We are publishing it because the negative result is the useful part.
Why it died: price trends through the bands, so fades get run over, and M5 costs eat what is left. Cost-dominated. Do not fund it — we did not.
Meta-Label Filter Research
A classifier sitting on top of the rule-based strategies, trained to keep only the entries resembling historical winners. This is what "the model filters most setups out" means.
Honest limits: 331 samples is a small training set and validation accuracy is not profitability. Treated as research, not as a finished product.