Deviating cell counts after bam subset

[Onero23]

Hi,
The tool is working fine however I don’t understand the results I get.
If I do a testrun with a 10x dataset (bam-file) and I use the detected barcode list (filtered_feature_bc_matrix barcodes) for a “subset” there should not be any data deleted in the bamfile (the bamfile should NOT be subsetted?). However if I use the bamtofastq tool afterwards and I use the cellranger count pipeline again on the new generated fastq files I loose many cells. (from around 3500 cells to 1800 cells). In addition if I repeat this I get different results (from 3500 cells to 1900cells). Can you give me any advice?
Thank you so much

Best

Wolfgang