Maximize Sharpe ratio on a single XGBClassifier trained on a 70/30 train/test holdout. Features centre on RSI oversold/overbought zones (RSI<30 buy, RSI>70 sell) reinforced by Bollinger Band %B, MACD histogram, Stochastic K, z-score mean reversion, and candle-body shape. XGBoost hyperparameters use moderate depth (4), a low learning rate (0.03), and L1/L2 regularisation to reduce overfitting and improve out-of-sample Sharpe. SL=0.30%, TP=0.60% gives a 1:2 R/R that supports po