One more update on this @Adam_Rivers and @Nicholas_Bokulich (sorry if this is getting spammy).
I tried the reverse read input strategy to qiime import, then itsxpress with the Taylor dataset this morning, and it seems to work. Except the read quality tails so badly in that dataset only ~1300 sequences are extracted.
Trying again with the test data I sent you yesterday, @Adam_Rivers, we do indeed end up with reasonable results with the reverse input strategy. demux-ITS.qzv (292.1 KB) 96% of reads pass (and the fails are probably at merging based on the BBmerge logs). Again, I think this should result in main strand, fwd orientation, so I would think no problem for parsing (shouldn't the tool essentially be blind to this?), but please let me know if you see otherwise.
As for why I was not getting output with this strategy yesterday...
It appears you are correct. Running with ITS2 selected results in output, whereas ALL gives no output. Thanks for pointing that out, but why? I would have expected just reporting ITS1 as missing rather than failing entirely... but now that I know this is fine.
Anyways, thanks for all the pointers on this. I'll work the reverse input reads through some additional analysis in the coming days .