Sharing Fide's approach to doing embedded evaluations. Would love feedback! And for those who want to collaborate, reach out to alex@fideai.org