Looks like a very outstanding engineering approach.
Different construction of the same tube type will have different musical characteristics, sometimes dramatically so especially when overdriven beyond linear operation. Even though measured performance would be equivalent in things like power supplies or most other non-audio circuitry, or even non-overdriven audio.
There is a unique difference when plugging guitars straight in versus using a common pedal or preamp of any type. The very limited magnetically generated power coming out of the quite high impedance of your pickups is directly connected through minimal resistance to the input pin of the even higher impedance tube electrode which extends into the vacuum. From that point on, your music is being basically sprayed through space.
So plugging direct (or true bypass) is the only way to be magnetically coupled directly to the vacuum you will be spraying your music through to begin with. You can feel it when you crank it whether overdriven or not.
A pedal or preamp inserted between guitar and legendary amp can overdrive the amp in its legendary way much easier sometimes because of the gain provided by the additional circuitry upstream of the amp, it can be very tasty but you are no longer magnetically coupled to a legendary tube amp.
Also unless the preamp/pedal is a simple clean gain stage such as an EH LPB2, which simply boosts the high impedance guitar signal unchanged using 9V battery power to give the amp more input amplitude than a plain guitar, the pedal is more likely to be generating the audible distortion rather than, or at least in addition to, the overdriven tubes.
You probably have more than one 12AX7, even most individual amps. Might as well try each one in the most critical input socket to get the most musical one picked out, along with checking for hum, hiss & microphonics which you don't want being amplified from that early of a gain stage.
It's only the big power tubes that can require matched pairs or quads, 12AX7's can be switched at will.
If some of your 12AX7's don't sound better than others, you've got all crummy tubes or there is something else concealing the full tone available.
Seems to me an emulator based on ML could be trained by live playing of the exact guitarist, guitar, and rig that you wanted to generate sound similar to.
Then it should be able to convincingly generate a synthetic tune of its own having the characteristic musical style of the player using the tone of the combined gear. Probably easier said than done.
Instead when you trigger tone output from the model digitally you would think some of the fingering variations might be heard too, but probably not any musical style, even if all are in the model to an extent.
When you plug in your guitar and play through it I think your own musical style would prevail like normal, and the fingering influence could be good for augmenting less expressive fretwork, but there are going to be times when you want to plug your own select guitar into just the amp portion of the model.
To get this option you would need to have sampling/insertion points between the elements, or at least between the guitar & amp.
You need to know what the signal looks like right before it goes into the amp so you can see the exclusive effect of the amp model alone and compensate for the guitarist & guitar (& pedal) during training.
If you're not going to draw on the musical style or instrument being in the model, maybe it shouldn't always be there to begin with.
It might be possible to switch them in or out like an effects loop.
Otherwise when you plug in your guitar and process it using the model, you're hearing a guitar plugged into another guitar plugged into a tube amp. And that's provided the training guitarist's musical style is only lurking silently.