I'm trying to model data with a time-homogenous CTMC with a number of states with corresponding constant transition rates $\lambda_{i}$ when I notice that much of the transition times from one state to another (lets call it $a \to b$) don't occur past a certain time, lets call it $t_0$.
I'm aware that much of the structure of a Continuous time Markov Chain is dependant on the memorylessness of the exponential distribution but I'm wondering if introducing location parameters into the distribution of first hitting times would interfere with that structure. How would the Kolmogorov equations ($\frac{dX(t)}{dt}=G X(t)$)of the process change? Can this be accounted for by making $\lambda_{a \to b} = \lambda(t)_{a\to b}$ ie. moving to a time inhomogeneous setting?
Whats the best way to address this without wholesale abandoning the CTMC framework?