From how you are describing your data (sorry, I still don't see pictures), your data is piecewise constant. At any given time step, you are comparing an established constant, to a hypothetical new constant of the last N values sampled. The object being that this new constant better explains the last N values than the currently accepted (long term) constant. Part of the problem, is figuring out what N is. As I have described it, this is a spline, and the problem with N is automatic (or variable) knot placement. I have seen some papers on selecting between piecewise constant and piecewise linear splines with variable knot placement, but off the top of my head I don't have URLs for you.