We have had issues with the replica GUIDs on and off for several months now. Historically, our work around was to have our users perform a hard close of the Collector App on their iPad mini 4, re-launch the app and sync. That process worked for about 2 years but recently performing the hard close then attempting to sync has led to data loss and we are still investigating the issue.
A common reason we used to see these replica errors was tied to a user performing an IoS update while they had an offline map downloaded on the device which was fairly easy to train users on proper sequence. Something about performing the IoS update seemed to create an issue with the SQLite database on the device and cause the GUID on the device to be changed because we could verify the lineage of the server GUID value and tie the server value back to the enterprise sde table.
As for the item not found error, we began seeing this sometime around 4+ months ago and we associate it roughly to the 18.0 timeframe of Collector for IoS. After getting some escalation on a ticket with ESRI and all of us being stumped we were getting off the call and I did a removal of an offline map from my personal iPhone. Presto the error was logged after we had spent nearly an hour trying to force the issue to be repeatable. The ESRI rep and I watched as I performed the test 4 more times of removing the offline map from the device and like clockwork that action would trigger the error to occur. We use SQL Server instead of Oracle so we do not get the extra Oracle text but otherwise the message is identical. Should be something fairly easy to test and confirm in your environment. We simply added this to our list of expected errors based on some of our workflows and put it alongside the errors that occur and say: Unable to process request. For input string: "18446744073709551558" which are expected and tied to the use of attachments.
