Smooth data prone to bifurcation

I have data that I am interrogating for certain trends. Sometimes, it's prone to bifurcation due to the bounds of the data. What I mean, is that the criteria I'm placing upon it will push it toward a boundary, so the result will move (bifurcate) to another solution, but then the next solution may remain in bounds, and so on. I've tried to provide some representative examples in the plot.

I want a solution that minimizes these large jumps. In the example, the green line would stay low until a little after 11 on the x-axis, and then go up and the purple line would simply stay high. With the yellow line, it's hard to say whether it would stay high or go low and then only go high around 12. It would be nice if somehow I could choose on lines like the yellow one. I have tried, smooth, medfilt1, and filloutliers, but the results were not as smooth as these examples suggest they could be. I'm aware of findpeaks, but try not to use it except for one-offs because it can be difficult in my experience to dial in the parameters.