Thanks for sharing that info, @elise_nghalipo!
Let's revisit the original error you shared, first (when running qiime taxa barplot):
This visualizer is telling us that it has no way to associate the feature IDs in your FeatureData[Taxonomy] Artifact with the feature IDs in your FeatureTable[Frequency].
So the next step is to figure out why (hint: this is related to the provenance graph I showed above!).
Let's take a closer look at "Step 4":
According to the earlier steps in your script 97_rep-seqs_3.qz is actually from your reference sequences dataset (the data used to train the classifier in the first place). So I think this makes sense - you are trying to combine two completely different datasets here, and they don't line up.
I think you should revise Step 4 to point the --i-reads step to the FeatureData[Sequence] that is produced in whatever step made table_3.qza (did you use q2-dada2 or q2-deblur)?
Also as a brief note. Filenames don't matter to QIIME 2 - you get to pick the filename, all QIIME 2 cares about is the semantic type of the Artifact. With that in mind though, I think you have a typo in "Step 2" which might've lead to this confusion. In Step 2 you named your output reference reads 97_rep-seqs_3.qza. Usually though "rep" means "representative" - as in "these sequences are representative of the samples in my study" - "ref" is shorthand for "reference - as in "these sequences are the reference sequences in which I will compare my representative sequences to."
Hope that helps!
:qiime2: