How Do You Determine Whether a Trading Strategy's Rules Are Robust to Small Data Differences?

Different historical datasets may contain slightly different prices or candle values. A strategy that changes dramatically because of a small data difference may be difficult to reproduce. How would you test the method's sensitivity to such differences, and what level of variation would concern you?