A dedicated paper-trading runtime and a realistic backtest engine were built during this period, explicitly modeling transaction costs, slippage and execution delay rather than assuming ideal, frictionless fills. This was a deliberate choice to make early evaluation numbers harder to fool, not easier.