QuantX is built the way a trading system should be built: capital preservation first, returns second. The risk engine is not a module the strategy can opt out of – position sizing, per-trade stops, portfolio exposure ceilings, daily drawdown limits and a hard kill switch are enforced above the strategy layer, so a misbehaving model cannot escalate its own risk budget.
The prediction layer combines gradient-boosted trees with a sequence model over multi-timeframe features, gated by a regime detector so strategies stay flat in conditions they were not trained for. The pipeline is audited for look-ahead bias and leakage, because a backtest that quietly reads the future is worse than no backtest.
Every stage is separable: research, backtest, paper trade, then semi-autonomous live execution with a human able to intervene. No martingale, no grid averaging, no unbounded position adding. This is research and execution infrastructure – it makes no claim about returns, and any deployment carries real risk of loss.
What it changed
- Risk controls enforced at the engine level rather than left to strategy code
- Explicit guards against look-ahead bias and data leakage in the feature pipeline
- Separable stages so nothing reaches live capital before paper trading