The Reality Gap in Small-Cap Backtesting