Expect a gap and try to explain it rather than excuse it. The usual contributors are execution costs modelled optimistically, fills assumed at prices that were never available in size, survivorship in the instrument universe, look-ahead in the data, and the trades you did not actually take because you were away, hesitant or tilted.
Quantify each. Compare realised entry and exit slippage against the assumed slippage-budget; count executed signals against generated ones, which frequently reveals a 10-20% shortfall in discretionary implementation; and compare the live r-distribution with the backtested one to see whether the shape changed or only the level.
A gap of 10-30% of expected return is normal and survivable. A live result that is negative when the backtest was strongly positive is not a costs problem - it is evidence the backtest was fitted, and more discipline will not repair it.
Related: incubation-period, slippage-budget, backtesting, performance-attribution