While you have some good points about the benefits of simulation vs a mathematical model, I think you missing the point with regards to the practicalities that the AMR team have to face when determining what features they develop.
The AMR team is only a small team of 3, and from what I understand, only one of them is the developer who implements the back-end optimisation algorithms. Also, I’m not even sure if the income from the subscriptions would allow them to work on AMR as their full-time job, I would imagine this is more of a side-hustle and they have actual day jobs, and most of the money from subscriptions would go towards paying for the server-time needed to host the website and run the optimisations and simulations for both current and classic wow.
Given there is a new expansion coming out later this year, I would presume most of their development time at the moment is focused on getting AMR prepared for Shadowlands, and they have mentioned as such in other posts on the forum.
So they don’t exactly have an unlimited amount of development time. They wrote a blog post a while back when the optimiser for Classic wow was introduced, explaining the reasoning behind the single one-off payment for classic. https://blog.askmrrobot.com/classic-pricing/.
I don’t know the proportion of AMR users who use the classic vs current optimisers, but I would think it is probably much more weighted towards current wow. So they only would have a relatively small budget to go towards the classic wow optimiser development.
The theorycrafting behind classic wow is a lot simpler than current wow, and has been well worked out for a long time now. On the other hand, there are many more complexities with current WoW, with a lot of subtle interactions between the various stats, talents, spells and gear, that developing a mathematical model that covers everything would be incredibly hard, if not impossible, so a simulation engine is required to be able to solve the optimisation problems.
Implementing a mathematical model is a lot easier than developing a whole new simulation engine, which they would have needed to do since a lot has changed between classic wow and current wow. And as Swol and yellowfive have mentioned, the mathematical model is actually quite a bit more flexible than a simulation model, and faster to compute, so given the amount of development time they had to create the optimiser, I think they made the right choice.
If they had unlimited funds and time to develop the classic wow optimiser, I’m sure they would love to add a whole lot of new features to it, including the ones you have mentioned, but unfortunately that is not practical. Being a mathematician and software developer myself, I can sympathise with the difficulty they have with trying to balance adding new features to the optimisers vs the practicalities of the amount of time and money they have available to develop AMR.