What your evaluating is not what you think it is. You're evaluating the models ability to execute multiple complex steps (think about all of the steps it takes

What your evaluating is not what you think it is. You're evaluating the models ability to execute multiple complex steps (think about all of the steps it takes for your second example) not so much if it is capable of doing those things. If you broke it down into 2-3 different prompts it could do all of those things easy.