There are several approaches to modeling,emulation, simulation
One approach uses tools like SPICE. It can start with an actual schematic of the amplifier. Spice simulates the component. Each component is measured separately and a component model put in a library. These are checked against each other. Make a circuit with physical circuits and run a wave through it. Make a circuit in spice and run a wave through it. These measure the same, and can null each other out. At different levels. At overload. Artifacts like hiss are often taken out to improve the model over the actual hardware. IK Multimedia copied the schematic of an Ampeg amp and during trying out the model it "Shutdown" when they cranked it. It turns out to be a design flaw in the Amp. In real life many designers are going to go to Spice or other software tools during their design. They can test before building a prototype.
Got back to the old way. A book with a bunch of reference pages, and look up tables. Spend a lot of time thinking through each part of the circuit stages. Figure out where you can insert points to measure the prototype to assure it is working as designed. Build a prototype, measure and test it, repeat each time you have it wrong. Build 20 prototypes because components have plus or minus 15-20% variances in some cases. If one in 20 breaks go back and design some more. It takes the same smarts either way. It just that much of the process is enhanced with software.
The designers are hardware engineers. After the complete amp is modeled they are compared level for level, by both ear and lab instruments. Some of the ear testing is done by musicians who have a different perspective on how it sounds.
Another copy/clone/profile method takes the actual output and does analysis. Then duplicates, tests, to figure out how to get the same signal. At many levels. This blackbox approach is what is convincing a lot of musicians to use modeling. It can copy their own amp and a level they like to play at. Their own amp may be flawed, and their profile of this amp is flawed, but they can only believe modeling if it can copy their own flawed amp. Later though they can try a golden model and hear what the original designer intended with the actual amp.
A lot of people here are not software and computer systems experts. Yet they are using technology to "talk" around the world. Take a little of that skill and try a demo of a modeler. It's not expensive. It does take some smarts to pick low latency software drivers, but it simple enough that must musician can figure it out. Some will fail with modeling, but it's usually something they're not following in how to use it. The end up forever blaming the modeling for not working.
As far as amp speaker interaction.
There is a feedback path from the speaker back through the amp, but the speaker is mostly a voltage follower. There isn't much interaction. Some even claim that zero feedback sounds the best.
The full feedback is from the musicians ear back through the whole rig. Things can be changed and the musician's brain makes adjustments. Which may include turning a knob. The real interaction is from the musician. Even if they're a DJ playing records.
Good luck in your search
