Olivier Nomblot
Black Eagle Neural is a genuinely multi-layered system: at its core run two neural networks—a primary net that reads sixteen engineered features (multi-horizon momentum, RSI, moving-average distance and slope, a volatility ratio, range position, candle structure, a volatility z-score, cyclical time-of-day encoding, the H1 higher-timeframe trend, and a EURUSD dollar-strength context) to call direction, and a second "meta" net that independently judges whether each trade is worth taking. It learns continuously—from the real outcome of every closed trade and, through shadow learning, from a virtual trade on every single bar—using triple-barrier labels that capture genuine take-profit/stop/timeout outcomes rather than naive next-bar guesses. Crucially, it polices itself: it runs out-of-sample walk-forward validation with purge-and-embargo gaps to block look-ahead leakage, keeps a snapshot of its best-validated weights, and a drift guard benches it the moment its measured edge collapses. On top sits a layer of judgment—Bayesian win-rate tracking, skill-calibrated confidence, regime- and session-level accuracy memory with auto-sidelining, and a cost-aware gate that demands the signal's edge exceed spread plus costs before it acts. Risk is fully engineered (ATR-based stops/targets, breakeven, trailing, an equity-drawdown circuit breaker, daily loss limit, optional news and daily-close controls, with fractional and meta-scaled position sizing), execution is hardened to MQL5-Market grade (ECN-safe order handling, stop/freeze-level compliance, retry logic), and the whole learned "brain" persists across sessions with a one-touch reset—all surfaced live on an on-chart dashboard. In short: a self-learning, self-validating, self-regulating decision engine, not a fixed rule set.
1
