Last Writer Wins is not the most compelling use case for CRDT because if those fruits were denormalized into Postgres rows, you'd already have the behavior desc

Last Writer Wins is not the most compelling use case for CRDT because if those fruits were denormalized into Postgres rows, you'd already have the behavior described. A more interesting example, might be if both users were inserting a new fruit, one at the beginning and one at the end. What should the sequence of fruits be after that? Likely a list with both new fruits included. But it depends on the domain and use case. In general, Sequence CRDTs seem to be the most difficult and most interesting area of continuous research. In my own work, I've implemented a sequence CRDT to create a distributed log of events across users that then (because the order is guaranteed to eventually be the same for each device) can just be reduced to whatever data structure you need. It means, we end up with one CRDT to rule them all and then can just write the "conflict resolution" in simple term with the business logic.