"Push to editable" copies a regular dataset to an editable dataset while keeping changes.
The first time you run a Push to editable recipe, it will simply copy the whole content of the regular dataset to the editable dataset. If you make changes to the content in the editable dataset, and then rerun the push to editable recipe, it will copy over all data that was new or changed in the original dataset but will preserve every modification you did in the editable dataset.
To identify what is considered as "new" or "was modified in editable dataset", you need to select some columns that form an identifier.
The main use case for a push to editable recipe is if you want to make some corrections to a dataset. For example, you have an input dataset of product categories in a database, but there are some errors inside, and for some reason, you can't get the error to be fixed in the source data: you use a push to editable recipe, fix the erroneous entries, and base the rest of the flow on the editable dataset.
Note that since editable datasets are limited to 100K rows, so are push to editable recipes.