When the merged changeset takes less journal chunks than the original one, the additional chunks remained there causing malformed or wrong data. Deleting merged changeset explicity before re-adding it will get rid of possibly leftover chunks to remain in journal DB.