Hi! Datasets in Prodigy are append-only by design, so you'll never lose a datapoint you collected by modifying it in place. But you can always load your collected annotations back into Prodigy by loading from an existing dataset instead of a file. See here: https://prodi.gy/docs/api-loaders#datasets Alternatively, you can also export your annotations using db-out and load them back in from a file.
Prodigy's input and output formats are the same, so you'll see the existing annotations and are able to correct them if needed. Make sure to save the results to a new dataset, so you don't end up with conflicts – and if you make a mistake, you can always go back to the previous state.