I was reading this article regarding controller action return types, and these two caught my eye.

1.

[HttpGet("{id}")]
[ProducesResponseType<Product>(StatusCodes.Status200OK)]
[ProducesResponseType(StatusCodes.Status404NotFound)]
public IResult GetById(int id)
{
    var product = _productContext.Products.Find(id);
    return product == null ? Results.NotFound() : Results.Ok(product);
}

[HttpGet("{id}")]
public Results<NotFound, Ok<Product>> GetById(int id)
{
    var product = _productContext.Products.Find(id);
    return product == null ? TypedResults.NotFound() : TypedResults.Ok(product);
}

Running the sample code, I noticed that for the second method, the possible responses and schemas are missing in Swagger UI, just a single 200 OK. However, I can see everything just fine on the first method. Only after adding [ProducesResponseType] attributes on the second method I can see the possible responses and schemas.

At the same time, the article says that with the Results<TResult1, TResultN> approach (second method):

All the [ProducesResponseType] attributes can be excluded, since the HttpResult implementation contributes automatically to the endpoint metadata.

When multiple IResult return types are needed, returning Results<TResult1, TResultN> is preferred over returning IResult. Returning Results<TResult1, TResultN> is preferred because generic union types automatically retain the endpoint metadata.

What am I missing here? Am I misunderstanding the article?