Most vst use modeling, it's far more then only sample playback, but you need authentic sound at base with layers and round robin, and not only relying on EQ DSP based and layer sound added at all the time which is obvious can make unwanted sound or things is not right. That why when people imports sound in a module, it does not play 100% the same, and without sufficient amount of layers and round robin, DSP will never replace native audio sound and retaining the real sound for authentic sound dynamic.
Roland have all physical modeling on playability already roll, swell, position sensing, buzz, but struggling on sound without sufficient of layers and round robin at base and multiple DSP modeling option + fx make the sound different then original recording.. Roland use 50% CPU stated from Micheal shack lately to get top notch hihat, and that why alesis strike hihat or low processor comparable module will never play nice under 2000$ modules with great low latency, if you want all editing and all feature and that play descent, you wont get that with cheap stuff.
If you want real drum playing, you need tip stick angle position detection on head-cymbal + on all surface area on triggers, on toms, snare, cymbal like a real drum, far beyond sensor that are dated. At the moment, Roland attempts to do roll, antis machine gun algo, buzz on toms\snare etc., it does not works great or excellent, you need all these base layers and round robin with high quality recording for descent dynamic and sound alternation on each hit. + all the power resource rest that run the module. You can do modeling, but you still need all high class audio sound recording, enough layers and round robin at base, 24 bit samples, you will never have good result under 2000$ module that play and sound descent with modeling, it will sound just ok to bad.
As far as mimic, there is no limit on layers and round robin how it's built, the limitation is the recording it self, there is swell, hihat that perform smooth as silk with high pressure sound on high pressure hihat, what are missing are position sensing on toms, cymbal, snare, that will add new level, but remain detection on tip stick angle on each hit on the surface position, that what make a pro drummer vs beginner or average, that require new triggers tech that don't exist and not sure it will ever, will require too much power to run these, far beyond Roland digital hype to prevent DIY triggering.
1 to 127 \ 128 is midi notes, 2box use 127 layers max sound that reach 127 midi note. The mimic "4095" is not sample", mimic don't have limit on layers and round robin, it's depend on the recording it self, mimic do not use midi, 1-127 inside with triggering. All computer software, use MIDI notes and are locked to 127, limited from the midi drivers on computer.
Infinite layers and round robin that is BS, no human can play, listen and record these, and it's all DSP based, it won't sound\ play accurately sound depth, dynamic from realistic recording sound on velocity. The hole story on modeling are from keyboard synthesizer, mix them with drums, it will just alternate the original sound like they do in a synth, if it's the attempts to recreate real drums it won't be the same as the authentic sound, record 8 different head, 4 tuning range, you can't do this with DSP, it will not sound as original recording side by side, that why people buy multiple VST with different studio, different mic, and different room recorded, it just works and it just sound real on album and organic as raw.
The modeling is very hard to operate, too much option that change the sound, you will find your self changing many things, but can add odd sound when another option is added, ish, ing, ting sound etc.. you will cycle back and forth like crazy, average users won't ever able to use it correctly, on module it show how much hard it is with Roland, what users want are sound great out of the box, many people don't understand threshold, modeling is far light years above it.