Hi! A few things to check to get a better idea of what could be happening:
- Are you using a workflow or recipe that generates new examples, e.g. by splitting sentences, like in
ner.correct
? If you have 60k examples with 2 sentences each on average, you'll end up with ~120k examples in total. So the number of annotations doesn't necessarily map to the number of incoming examples. - Can you try upgrading to the newest version? We also have an alpha version of the upcoming release that includes some fixes around multi-user workflows that you could try: Duplicate annotations in output - #12 by kab
- There can be effects where Prodigy re-sends a batch if an annotator requested a batch and never submitted it, and it expired. This "work stealing" is important so you don't lose any data if an annotator stops working and doesn't save in a given time period. That said, 1k examples difference would be a lot here and this would normally only affect a batch or two.