We're experimenting with use of sync-enabled feature services to support a mobile application. We're using archiving as opposed to versioning to detect changes. In a simple test I added a new feature to the parent feature class in the enterprise geodatabase to see if it would sync down to the client (a runtime geodatabase on a laptop). Eventually the new feature showed up on the client -- which is good. However, I noticed that ALL of the features in the runtime gdb had the same gdb_from_date value as my new feature. I would have expected my new feature to have a layer gdb_from_date than all other records. Which prompts my question, what/how much information is actually transferred to the client on a "sync"? I thought it would be only the changes since the last "create replica" or "sync replica" operation. But it seems like it might be re-sending the entire dataset.
Any insights would be much appreciated.