Hello @jose_gacia
Here are my suggestion, I think that you are getting low resolution because your classifier isn’t doing a great job of representing your sequences. You could try to find a better one and that might help you get a better taxonomy result. One that might work is : SILVA 138 DB available on the 2020.8 data resources page. It has better rank prefixes (i.e. d__ instead of D_0__ , etc…) and is generally a smaller database.
However, if you are getting successful results by running blast on the rep-seq.qza sequences that didn’t get identified that is awesome and a great strategy. There are a couple of things that could make this easier. 1) you might be able to download the blast db and query against it locally which could save time compared to clicking sequences individually. 2) you also could look into the command classify-consensus blast. Either way I think you strategy is good. Hopefully my suggestions can help with saving some time.
Finally you asked what the frequency from the table.qzv means. The frequency per sample looks at how many features are in each sample and frequency per feature which looks at how many samples contain that feature.
I hope this helps!