The confidence interval spanning zero means the model could actually be unprofitable once deployed. - Backtest ROI may overstate live performance due to execution differences (getting worse odds, missed bets, etc.). - Model selection bias (you chose this model because it looked good in the backtest)