Hi Ollie,
Just to close the loop on this.
At the outset, I wanted to say IMHO Toontrack/EZd/SD are GREAT products. I'm doing various projects currently with my similarly middle-aged muso friends, including one live project where I sampled-replaced the entire kit of our drummer with the SD Avatar sample library. He has a double bass drum set up, 5 toms, 2 rides, umpteen crashes etc. etc.... ..and a cowbell. He's a very busy player (needs to be with this kit!) and I sample-replaced his entire drum part based on 4 channels of original drums (kick, snare, 2x overheads). No one noticed, including the keyboard player who claims to be able to spot sampled drums at 1,000 yards distance... In fact, every one said how great the drums sounded! So well done Toontrack!
Anyway, back to MHE.
As you suggested, I tried this using MHE maxed out and an 8 voice limit on the snare (default setting) and using the Americana kit which I happened to have loaded up in Superior Drummer. I used a straight backbeat snare at 144bpm and constant midi velocity of 115. As far as I could tell - based on listening and inspection of the waveform - this made no difference at all (with all humanisation ON except for my bugbear Semi Seq).
I then tried it again on a fast semi-quaver snare fill. This was made up of a series of midi snare hits of different velocities. Otherwise, same as above (Americana kit in SD, 8 voice limit, all humanisation ON except for Semi Seq). This time there was a small difference. On close inspection, it turned out that 2 of the midi hits were adjacent and the same 104 velocity. MHE seemed to changed these a little bit.
So, my experience seems to be that MHE on heads only works for midi hits of the same velocity that are close together. And it's therefore not the same as recording a series of "non-cold start" snare hits and then slicing them up.
Finally, back to the question of how SD generates its samples on playback (issue number 1 in my original post). With humanisation set to "Random + Vel to Vol + Soft Vel" I basically just get one sample - the same one - played back repeatedly for a back-beat snare. I don't quite understand why, if there are15-20 (whatever the number is) of samples per velocity layer, SD doesn't play back a more random series of samples? But it doesn't. Just one.
If I then switch on "Alt" in humanisation as well, I get 2 samples alternating, and sounding a bit "tick tock". Again, why only 2? Seems odd. What waste of the other samples in the layer!
Finally, if I go to full all on humanisation (everything on, including Semi Seq), I get some extra samples, which seem to include "hard hit" layer samples (noticeably louder and harsher timbre) and some noticeably softer (volume and timbre) samples from lower gradient hit layers. As mentioned, this makes the drums very hard to sit in the mix.
So, my basic point remains - how to get SD to play those 15-20 samples in the relevant velocity layer randomly, rather than just one or two? As far as I can tell, SD doesn't really work in this regard. At least, not as I would have expected it to, based on what it says in the manual.
As mentioned, you have a great product. And I've found a way of working around this using audio bounces. Plus I actually don't mind if back beat snares and the kick are identical - I think in a mix you get some psycho-acoustic effects which fox all but diligent and experienced listeners into thinking the hits are subtly different.
Anyway, this started out as some feedback for TT from just one happy used on how to improve your product.
Rgds,
Mark