When you're evaluating a decision, you have partial information about the extremes. If you've been with someone for a year, you know how bad your worst fight was. Chances are though, whatever it was is a sampling from a distribution that you've only sampled a small amount of.
Your eventual worst fight will be worse, because when you have more samples chances are the extremes will be more extreme. (BTW if you're evaluating algorithmic trading strategies you need to think about this.)
If you're already thinking things are bad enough to consider leaving, you can look forward to them getting worse even if the underlying dynamics are no different.
Note the asymmetry and the importance of the extremes. Everyone who has a breakup story also had good moments and may have mostly enjoyed their average day. But you choose to leave because of the lows, not the average or the highs.