Should I use -a or -g when demultiplexing ONT reads with dual barcodes?

cutadapt 4.9

I have 16S amplicon reads that were sequenced with ONT that I am trying to demultiplex. Each sample was PCR barcoded with a 13 base barcode on both ends, so I expect a read to start with a barcode and end with its reverse complement. I put together a fasta file of all my pairs, some are listed below.

>HL001_FW
ATCCGGTCGGAGA...TCTCCGACCGGAT
>HL002_FW
CTGAGGTGATCAG...CTGATCACCTCAG
>HL003_FW
AGTGTCCTGCTAG...CTAGCAGGACACT
>HL004_FW
ATAAGCAATTCGA...TCGAATTGCTTAT

The problem I run into is whether to use the -a or -g flag. Looking through the documentation I see it used almost interchangeably for linked adapters, but I get different outputs depending on which I use and I'm not sure which is correct. I used the below commands, for reference

cutadapt -e 1 -a file:barcodes_for_cutadapt.fasta -o trimmed-{name}.fastq.gz reads.fastq.gz

cutadapt -e 1 -g file:barcodes_for_cutadapt.fasta -o trimmed-{name}.fastq.gz reads.fastq.gz

Start with the cutadapt documentation for the -a and -g options and linked adapters, then reproduce both command lines using the supplied barcode FASTA structure. Compare the demultiplexing results and determine what guidance the documentation should provide about choosing between the two options.