Exactly! This used to trip me up every time I thought about it, but the key is that the barcodes aren't in the sequences themselves if you have the I1/I2 files. So nothing needs to be searched within your reads.
The QIIME 2 demux implementation will discard anything that doesn't match the barcodes it received.
How does your Per-sample sequence counts table in your viz look (its on the first tab, scroll down)? Are you missing any samples?
P.S. you can attach .qza/.qzv files to the forum as well (as long as they aren't too big).