Stocks that jump at the open on hype usually give some of it back during the day. The bot bets on that pull-back by shorting them (selling borrowed shares and buying them back cheaper). Every position has an automatic stop that caps the loss.

Paper = real orders on TradeZero with play money: real fills, real borrow checks. This is the real test. Simulation = the bot replaying each day's actual prices after the close, assuming perfect fills. If the two drift apart, reality is costing more than the backtest assumed.