Two separate choices
You can say a result looks wrong without allowing your text to be used for training. The optional training toggle starts off.
Keep provenance
A future training pipeline needs the source, label, consent, and model version. User labels need review before becoming trusted training data.
