One optional addition could be a randomize amount factor, something like AMT.
With the factor set to zero, the track still outputs no more than N voices, but in a reproducable manner: simply using the first N pitch values stored in the step (if present). With the factor set to full, we get a full random behavior.
Now when playing my example C – E – G sequence, and randomise factor set to zero, it plays C – E – G in 100% of the cases. But opening the factor a little bit will introduce the alternative C – F – G sequence only now and then, like an occasional mutation. Opening the factor slowly to full results in a 50/50 chance in getting both variatons.