Hi @PatoUru,
I think closed reference gets a bad rep here. Closed reference gets a bad rep in general becaue the idea of (a) discarding sequences and (b) not having something robust to multiple enviroments is unattractive to many people.
And I think it depends. I've been playing wtih some benchmarks in this area myself. ...The classifier tends to give me more unclassified reads for my ASVs which makes identification more difficult and collapsing a challenge. Do you use a per-region classifier, a bespoke classifier, a full length classifier here? (Is it approrpiate to apply different methodology to different regions and studies?) Like, these are all questions you need to ask.
I also believe a recent thread talked about how fragment insertion doesn't work as well on mixed regions (so if you're combining V3, V34, etc it might not be as effective.)
...I'll also say there's potentially some benefit in being able to look at bray curtis distance which fragment insertion doesn't allow you to do.
Best,
Justine