One thing that makes this more complicated (and take this with a grain of salt, I'm making lots of assumptions): I think in many cases the axe doesn't need to simulate individual resistors and capacitors, it just needs to simulate their combined effect. To drastically simplify, if you have a computer function that does 2 + 5 + 3 + X, to optimize it you can just do 10 + X. Similarly, if you have a grid of resistors and capacitors that does some set of low and high cuts you can replace it with a simple function saying "low cut at X Hz, with Y slope, high cut at Z Hz with W slope". This makes the models more efficient, lower latency, and allows more effects and advanced parameters to run. But it also means that the tools you create can't expose "set value for this resistor" as easily as they can expose "set the transformer mismatch".
In fact, that's a perfect example: Transformer mismatch is a number that represents how the power tubes and output transformer interact. When you swap power tube types in an amp, part of the audio effect is maybe from the new tubes themselves, but most is because the new tube type won't be properly matched to the transformer anymore. But in the Axe, you don't need to set one transformer value to 234 to match the 6L6 value of 234, you just set it to 1.0 to mark it as evenly matched. And turn it up or down to simulate different levels of mismatch. This is because the actual values of the transformer don't matter, just how matched it is to the power tubes.
But consequently, you can't set a transformer to a specific value, or automatically mismatch you power tubes because those values aren't in the simulation. They've been simplified (in the mathematical way, like above when I combined constant values into a single one which is functionally identical) so now you just have a simple value where you can adjust the ratio of the match/mismatch.
Lots of the advanced parameters I'm sure are exposing some of these sorts of optimized values. But that's also what makes it difficult to make a true schematic builder at the resistor and capacitor level: You'd be losing all of that optimization.
An interesting idea would be a sort of compiler program though, where you could modify the input schematic then compile a model in the optimized form.
But TL;DR: I think because of the way the models are optimized to run in a real time, low latency way, there are many cases where an individual resistor value can't be tweaked because that entire chunk of circuitry has been baked into a single thing representing the result.