Introduced on day 14
Cluster stability
How consistently the same groups are recovered when the data are resampled, the algorithm is restarted or the period changes.
In detail
Typical checks refit on bootstrap resamples or subsamples and compare labellings with the adjusted Rand index, or track the Jaccard similarity of individual clusters. High stability shows repeatability, not meaning: a split on one wide column is stable too. Instability across time (label churn after a refit) is a practical reason not to attach standing offers or names to segments.