Offline support usually gets added late, as a cache. Then two devices edit the same record, one overwrites the other, and a user loses an hour of work. That is not a caching bug, it is a missing conflict model. Write queue before cache. Reading offline is easy, writing offline is the hard part. Every mutation made while offline has to become a durable queued operation with a client generated id, a timestamp and the entity version the user was looking at. In React Native that usually means SQLite or MMKV, in Flutter Drift or Isar. The storage choice matters far less than whether the queue survives a force quit. Idempotency keys, not retries alone. A queued write will be sent more than once. Send a client generated idempotency key with every mutation and have the server store the result against that key, so a repeat arrival returns the original result instead of charging the card twice. Pick a conflict rule and write it down. Last write wins is a real strategy as long as it is c