ner.manual - how to navigate back/forward through source sentences

I just started using Prodigy and I could not find out how to navigate back/forward through the source sentences.
Yes, it is possible to navigate back for current annotation session, but how to do it for the sentences that are stored/saved in previous annotation sessions?

Hi! Datasets in Prodigy are append-only by design so you never lose or overwrite any datapoints. The latest 10 examples are kept on the client before they're sent back to the server, so you can hit undo to correct a previous decision. If you want to revisit existing annotations saved in a dataset, you can load them back in using the dataset: prefix: Just make sure to save the result to a new dataset.

If you find yourself annotating and wanting to go back and forth a lot to check the context etc., this can often indicate that the segmentation isn't ideal for the task. For example, you might want to disable sentence segmentation and annotate whole paragraphs instead. It's also possible that there are improvements you can make to the label scheme so it's more intuitive to annotate.

I would like to clarify a few things here as I was looking for something similar. So let say I annotated a few samples and saved them. The moment I press save button the history vanishes.
Is there a way by which we can navigate to the previously annotated sentence/ example and change the label as per our wish? Also I want to navigate back to the examples annotated in the previous session.

Yes, sorry if this was confusing or unclear: in that scenario, you can restart the recipe with dataset:your_previous_dataset (instead of something like data.jsonl) as the input source. This will let you go through and re-annotate the existing annotations in the dataset again.

One thing to keep in mind here is that datasets are append-only, so new annotations will create a new datapoint (and you typically want to save them to a new dataset). The reason behind this is that you should always be able to go back to a given state, discard a dataset, try again, and reproduce what the annotator saw. For example, if annotator 1 labels some examples and you load them back later and make different decisions, you wouldn't want your decisions to overwrite the previous annotations. You may choose to use your new annotations, or resolve conflicts using a workflow like review, but by default, they'll be new datapoints.

Thanks. this is one way but what I want is that my annotator wants to see the previously annotated examples in the same session. And also want to change it if needed in the same session. So it will be in prod and once is started the annotator will be notified . As there are too many examples to annotate so he cannot annotate in one go. So he will save it and open the link next day and start annotating. But next day he wants to see his previously annotated examples.
Now everyday we don't want to start a new session. there is a section called history but that disappears after saving.