There is no way to know for a given collision. However, for a statistically large number of events, we can make a selection so that we enhance the amount of signal (the particle of interest) in it. In other words, only the presence of the particle could explain data. The challenge consists in designing the selection well enough.