Currently, when a user saves cluster memberships from a segmentation exercise (k-means, for example), the variable is dependent on the calculation. So if a user accidentally modifies the calculation, then the segment variable would recalculate.

This feature request suggests that users have the option to hardcode the segment membership variable when saved so that it is not dependent on the segmentation calculation.

The workaround would be to save the dataset and merge it (or upload it to the same document and create a relationship).