Hi! If you're annotating manually, you should ideally correct the annotations and hit accept, so that the answer you're submitting is the final and correct one. If that's not possible, e.g. because the tokenization doesn't match, there are OCR issues or other problems, you can hit reject to exclude the example from training and to separate it from the examples you skip for other reasons. See this thread for details:
Alternatively, as of Prodigy v1.10, you can also use the buttons setting in your config to disable the reject button entirely: Installation & Setup · Prodigy · An annotation tool for AI, Machine Learning & NLP