The point is, games where things are meant to be 50/50 usually record past events, and then work to normalise the distribution - i.e. each time you don't get para, it increases the probability of para next time, so that you are consistently drawn towards 50:50 on a micro scale, not over a sample...