@Orvillain I'm ignoring all the UX suggestions and minor glitches as I'm completely unsatisfied with how it behaves at the moment and I will have to do a third or fourth complete redesign of the UI of this thing.
The only glitch I see in this video is that if you have more than 2 groups (eg. 5 or 8) it still only performs 3 groups - probably it knows that there is no sample in the last group and then it resets the counter, but I'm not sure why it does that.
But in general it's expected that you use as much RR groups as you have RR variations, so you're already in weirdo land if you use more RR groups than samples.
Now if you have different amounts of round robin variations within a single samplemap you can achieve this by using multiple RR layers - which is where this complex system starts to show its strength, for simple RR stuff you don't need that in the first place.
So let's assume you have closed hi hats and open hi hats in the same sample map. For closed hi hats you have 8 RR variations, but for the open hi hat you only have 6 RR variations. You cannot setup a single RR layer to work with both types, instead you need to create 2 RR layers:
One for the closed hi hat samples with 8 groups
One for the open hi hat samples with 6 groups
And then we assign the closed hi hat samples to be cycled by the first layer and the open ones by the second. Now comes the important part: in order to tell HISE which samples are subject to which layer you can use the Ignore flag (so in this case both layers need to have the ignore flag enabled). This works like this:
Assign the closed hi hats to their respective group in the first layer
Assign the open hi hats to their respective group in the second layer
Assign all closed hi hat samples to the ignore group in the second layer
Assign all open hi hat samples to the ignore group in the first layer.
Note that this idea can be combined with any layer so you can create very complex group arrangements. If only the UX would make it easier to understand that stuff...