Backtest-Live Discrepancy

How significant has the difference been between backtested performance and live trading results? What concrete adjustments did you make once that gap became clear?