Every other validation method reuses history that you have already looked at. Incubation generates data that did not exist when the strategy was built, which is the one thing no amount of statistical care can manufacture.
Set the length and the criteria in advance: how many trades, over what period, hitting what minimum performance, with what maximum deviation from backtest expectations. Three to six months, or at least 30 to 50 trades, is a common minimum, and it is less than the statistics would demand.
Incubation also shakes out the unglamorous failures: a data feed that differs from research, a timezone bug, an order type the broker rejects, a signal that fires at a moment when the market is closed. Those are found in weeks and would be found expensively in live trading.
Related: paper-trading, out-of-sample, live-vs-backtest-divergence, strategy-monitoring