Occasionally I see table definitions that use an array of jsonb (i.e. `jsonb[]`) rather than a JSON value with an array. 

So why would one use

```
create table data
(
  ... other columns ...
  payload jsonb[] not null
);

insert into data (payload)
values 
(array['{"key1": "value1"}', '{"key2": "value2"}']);
```

over 

```
create table data
(
  ... other columns ...
  paylod jsonb not null
);

insert into data (payload)
values 
('[{"key1": "value1"}', '{"key2": "value2"}]');
```

As I see the first solution every now and then I start wondering if there is some upside to that solution compared to the "json only" solution that I don't see. 

With the second solution everything is "json", when accessing data, I don't have to mix native array operators/functions with json operators/functions. Which I think is a lot easier to deal with. 

Can anyone see a real benefit of mixing native arrays with jsonb rather than simply using JSON arrays? (Or any downside of **only** using `jsonb`)

Obviously this assumes that de-normalizing the model indeed makes sense to begin with.