I wouldn't be too hard on yourself, Dave...I have read many of your posts and you have a better grasp on this stuff than many and more importantly, you are open to learning...

Dynamic modeling would indeed be helpful, but even Porta noted that his Lempor theory, which was derived from basic physics, did not account for time-varying behavior. Often in engineering design we will build steady state models at different operating points (high load, low load, hot day, cold day, humid day, etc.) which help to bracket the operating regime. Components can then be sized to account for the most demanding conditions. The steady state calculations are much easier to solve than dynamic modeling equations which generally require "numerical methods" to produce a result - that is to say they are equations for which a definite answer cannot be obtained, but can be written in approximate forms that require a computer to "guess and check" until a solution is reached...the solution is technically not an exact number, but one with such a small error, that it can be counted for all practical purposes as exact.

Early in my career, I helped build out a dynamic system model. The software we used (Matlab/Simulink) allowed us to create a library of components (valves, heat exchangers, pumps, etc.) and then link them together to mathematically represent the prototype fuel cell power plant we were developing. It was far from a trivial effort and some models still required empirical relationships. I spent many months monitoring simulation runs to make sure they were giving reasonable results (as checked against the steady state models and lab testing). Hopefully someday someone will be able to build a similar model for a steam locomotive...

One of the challenges with modeling is picking the resolution of the model. We could build an incredibly complex model of cylinder phenomena or staybolt stresses and probably get some very useful insights into what was going on. That would require some very expensive software and experienced engineers to properly capture the physics and interpret the results. Those models would also be too complex to run with a complete multi-physics locomotive model. For example, you wouldn't build a whole boiler stress model to study staybolt stresses. You would want to concentrate on a smaller chunk of the firebox so you wouldn't have to wait days to get a result...although I suppose if you had access to a supercomputer, you might be able to get the job done ;)