The fact of the matter is that the grounds of majority of theories come into fashion but become decrepit like clothes because the researchers become more informed with time.
Dear Viktor, the answer to question is not simple, despite the fact that process has an orderly periodicity. Unpredictable new data and exponential, essentially nonlinear evolutionary cognitive processes do not allow to give long-term detailed forecasts, as well as to estimate the range of human cumulative knowledge wich affects by the changes and where, when the new "loop" of knowledge "cells collapse" begins.
IMHO absolute forecast can only be in the final "final theory", which is incompatible with cognitive process (i.e., not achievable, reality can not contain its absolute model) and being (i.e. existance as process at all), it is impossible theory, otherwise existance cathegory has a big problem.
Of course, for better understanding of how this happens it is necessary to begin with the classics: understanding dialectics and the fundamental work on the problem ("The Structure of Scientific Revolutions" by Thomas S. Kuhn).
Since the process has in its essence a physical nature, like everything in our reality, its roots are hidden in the stability of dynamic systems. So, it is also useful to have a general acquaintance with the theory of self-organized criticality. It is useful for better feel the dynamics of cognitive processes.
In any case, the process begins with the accumulation of a certain amount of knowledge (experience) that are inconsistent with previously known models. Contradiction generates in the case of traces to alterative patterns something like antinomy within the data currently available. Some time both knowledge systems exist in equals. This is phase of cognitive dissonance, as a rule, it ends with the fact that outdated (model) system finally loses its stability and crumbles. The system of knowledge, together with the entire physical structure of its carriers, as a dynamic system, falls into a new stable with respect to interferences (ie, new information) state. The more complex system of cumulative human knowledge, the more large-scale restructuring in it can occur.