I suspect you have hit the nail on the head here. Also when nodes relocate they should not delete chunks, but upload them to the network again. That can have this affect, same with splits. Nodes restarting shoudl also do this.
So a rule is a node never deletes data without first saving it again to the network. It’s a good rule, but can “hide” errors at this stage, however we will find those errors. I am unconvinced with anything short of full AE that these things will happen.
We have mostly AE in place but rushed this baby out to prove a lot more and that has been great, next one (soon) will be significantly better and on we go.