If we were asking about any generated output other than code, the answer would be an obvious and resounding "No, you can't judge a program based on its output."
And I know that's not exactly what's being asked here, but that's essentially the intent behind the first two scenarios described in this question. In these cases, we're stuck with a user who has put the cart before the wagon. He's written an app to generate output... but he's unsure what the output should look like.
So, realistically, what should be reviewed, is a sample of what the generator might produce.
If you're writing code-generation code, you need a very concise, specific goal as for what the generated code should look like. And I'm completely fine with seeing a project created for the sake of hammering out what the output of the generated code should look like. And in this case, that it is generated code is irrelevant (and should be left out of the question).
From here, once the template for the generated code has been reviewed and hammered down, and the developer creating the code generation software is satisfied with the output, then the code generation software should be based on its code--not on the code it outputs.
In no other circumstance would we consider it acceptable to review a piece of code based on its output--why should output that happens to itself be code be any different?