As a couple other comments pointed out, it's also not fair to judge Claude based on a one shot like this. I sort of assume these limitations will remain even if we went back and forth but to be fair, I didn't try that more than a few times in this investigation. Maybe on try three it totally nails it.

Very true, I would also caution this with test projects with real humans in the hiring process. Comparing one-shots from actual people is unfair too, and often the most valid assessment comes with giving them feedback and seeing how they respond to it.

Aside from that point: if you are reading this and making people do a project as part of the hiring process, you should absolutely be paying them for their time (even a token amount).