Thank you for the added detail!
There's a bunch of things that can mess up a database. For example, it was necessary to reformat taxonomy labels within Unite before they would match for Qiime2.
To build and test your database, you could try using the RESCRIPt pipeline or download one of my pre-trained classifiers for the UNITE database.
You could also try a totally different approach!
https://docs.qiime2.org/2023.5/plugins/available/feature-classifier/classify-consensus-vsearch/
Using classify-consensus-vsearch will also solve this problem, as it searches the database in both the forward and reverse directions by default!