Hi @bkramer ,
It does not look like the odd results have anything to do with the taxonomy classifier — it is classifying a single sequence because this appears to be the number of input sequences (judging from the table and rep seqs files that you supplied).
You have one ASV with a total frequency of 1, even though the number of sequences is much larger post-demux... seems quite likely that something went wrong during denoising, perhaps paired-end reads failed to merge? You should check out your dada2 denoising stats summary for clues.
Good luck!
P.S., on a side note:
Since you are using 515f/806r primers, you could use the pre-trained SILVA 138 (or 132) classifiers from the QIIME 2 website to save some hassle.