CUTS_
The cuts can be moved and essentially nothing changes group, and the density does not follow them. The partition is not doing arbitrary work: the components are divided by gaps the data actually has.
The informative reading, and an uncommon one. Over the designed experiment it arose on 10.3% of measurable fits and the component count was right on 99.0% of them, against 57.4% over all measurable fits. It holds within adequacy bands rather than merely tracking them — 96.8% within MARGINAL and 100% within AMPLE, on 218 and 511 attempts — so it is telling you something the sample size and separability do not.