REWARD
OVER RISK.

The edge · 1,601 trades, fees in

Cumulative R — every trade, in order

Why a 32.9% win rate is the point

The ensemble · 4 of 6 cells armed

Published in the open: we also tested mean reversion on 4h (profit factor 1.00, cut) and breakout on 4h (profit factor 1.09, marginal). Neither cleared the bar, so neither is armed. We'd rather show you what we cut than pretend everything works.

The honest part — in-sample vs out-of-sample

Any strategy can be curve-fit to look good on the data it was built on. To expose that, we split the timeline: the first 60% is the period the approach was designed around (in-sample), and the rest is data it never saw (out-of-sample). The out-of-sample profit factor held — and all 5 sequential walk-forward folds were profitable (2.17, 2.00, 1.12, 2.10, 1.11).

Methodology

Biggest wins — the fat tail the trail rides

The eight largest of 1,601 simulated trades. These are not a typical result — the average trade is +0.28R and a typical loser is −0.74R. The whole point of the engine is that rare winners like these, captured by letting the trail run, more than pay for the frequent small losses.

Frequently asked

Is this a live trading account?

No. These are backtested results: the published engine simulated on real Binance candles with no lookahead, entries filled at the next candle's open, and 0.2% round-trip costs deducted from every trade. The live site runs the identical engine code.

Why is the win rate 32.9% and not higher?

On purpose. We optimise reward-to-risk, not hit rate. The average winner is +2.35R (about +21.4%) while the average loser is −0.74R (about −6.5%) — a 3.2:1 ratio. At that ratio you only need 24% of trades to win to break even; at 32.9% the edge compounds. Frequent small losses and rare large winners are exactly how disciplined trend-following behaves. Anyone advertising an 80–90% win rate is selling the wrong number.

What does "R" mean?

R is the amount risked on a trade — the distance to the initial ATR-based stop. Measuring every outcome in R makes a 1-day Bitcoin trade and a 4-hour altcoin trade directly comparable, and it is leverage-free. "+0.28R expectancy" means the average trade returns 0.28× what it risked, after fees.

Why publish in-sample AND out-of-sample numbers?

Any strategy can be tuned to look good on the data it was designed around. Splitting the timeline and reporting the period the rules never saw (out-of-sample) is how we expose overfitting instead of hiding it. Here the out-of-sample profit factor (1.58) actually held above the in-sample one (1.55), and every one of the 5 walk-forward folds was profitable.

What are the four strategies?

An adaptive ensemble. A regime classifier sorts the market into trend, range or high-volatility, then arms only the strategy validated for that regime: trend (on the 1-day and 4-hour timeframes), mean-reversion (1-day) and breakout (1-day). Two more cells were tested and did not clear the bar — they are listed openly below rather than quietly dropped.

Will live signals perform like this?

Not necessarily. Backtests benefit from hindsight and assume idealised fills, and any window can flatter a strategy. Expect live results to vary — treat every signal as education, not financial advice.