inb4 anthropic classifiers start picking up this style of structured output as against policy too...
i'll never understand those things or why they hate structured output in chat results.
every time i see something flagged, i'm like... uh, no... this isn't your cot or me doing anything weird with the output data... this is a structured output requirement to help me understand states of things n junk. it's basically just... lazy man wants to glance at a standard form of generic questions that help explain the answer and how much to trust it without having to prompt it out of the model every time.
and it always makes me wonder... at what point will it refuse to answer the question of a basic "why?" to someone because they consider the response some sort of miraculous answer that must not be questioned for reasons behind it. reasons are ip for some reason... answers... ok, you can have those, unless it looks like rationale... sigh...
anyway.
grumble grumble... rant over.