In addition to the CP observation renumbering, I’ve also experienced a new wrinkle: CP dropping columns from the output file if requested features are not found in the initial segmentation.
This is a serious problem, not least because dropping columns forces a re-numbering of remaining columns, and results in a difference between column headers in the Output.CSV and the ColumnHeaders.CSV.
Overall, CP needs to do a better job of tracking object numbers. I’ve posted on FBOM renumbering observations (which makes downstream observations unreliable). Now, I’ve discovered that CP is dropping columns, wiping out whole variables if they are not present in a first segmentation. All objects discovered in initial segmentation must be preserved, and all requested data columns must be preserved, even if they don’t have initial values. Otherwise, working with the data post-facto becomes very difficult.