Let's not get carried away. There's very little factual evidence, some of which are under dispute (e.g., the last Inmarsat data point).
So there are many plausible scenarios -- with or without pilot intervention -- which can be made to fit the available data depending on how one chooses to interpret the data.
Also, even a Cat D simulator is not an engineering simulator. A Cat D simulator is only designed to provide fidelity for line operations within a specific flight envelope. Generally that means bank angles < 45 degrees, nose attitude no more than -10 deg down or +25 deg up, and airspeeds within the aircraft's normal operating range. Anything outside of that and you have to be very careful to cross-check against the simulator's validated flight envelope (e.g., using alpha-beta plots, etc.).
That is why a Cat D simulator can only provide limited stall / upset recognition and recovery training. Yet here we're talking about speeds exceeding Mach 1, 80 degree bank angles, 15,000 fpm descents, etc.
Those figures are well past the parameters of line simulators, and it would be a crap shoot whether the behavior seen have any semblance with reality or not.