For ONC 2015 certification in 2021, Patients have to be granted access to their data records due to the interoperable FHIR requirements. ONC has mandated that means access via FHIR to the USCDI data set. In their federal guidelines they indicate they will at some point mandate all access to a patients data set save for things like psychological notes or other things that would put the patient’s health at risk.
If you use FHIR resource types for the data transmission and package them in the ndjson format you should be able to make it compatible w/ the bulk $import process.
I agree as far as the the problems that arise when dealing with conflicts between source and target data. The FHIR format allows a resource to have multiple versions of a record and the status of those records can be ‘proposed’ and I think I even saw somewhere that some had a ‘contested’ flag. So there are ways to work through these things but it will definitely be a ‘fun’ engineering project.