Long-form fictional roleplay environments as AI evaluation sandboxes

> I would love to know whether OpenAI has considered using long-form fictional roleplay environments as evaluation sandboxes, or whether there is a better place to submit a more detailed version of this concept. This is a developer forum for those using OpenAI models. Your topic seems to be beyond the scope of what any OpenAI staff member would respond to.