If the documents are saved into one PDF file and the number of pages can vary, the assembly activity separates documents incorrectly.
This happens, because some pages could be incorrectly detected as the first page, and the document will be separated.
To avoid this problem, a separate classification skill should be created. It should have the first-page class and the other pages class for the particular document type:
This classification skill should be used in an assembly activity within the processing skill:
With this options your documents will be correctly assembled.