Running into an issue with visitor vs. hit level segments, that's also causing issues with virtual report suites and the new distinct count segment operator.
Basically, it seems like regardless of where segments are applied, visitor segment is always applied first, then hit-based segmentation second.
Visitor level segment where it only includes visitors that have 1 distinct value for something like "shopping category"
Apply a hit-level segment to exclude some category values (possibly due to accidentally bad data, or if you want to analyze a subset).
The result is you still have visitors in the 1 distinct value segment, but they have no hits. These visitors should be moved into the 0 distinct value segment since their hits were deleted.
The problem that exists currently is that if you end up with data pollution, you have no way to delete bad data and the dimension now becomes error-prone when doing the intended reporting.