@dotnet-policy-service agree

What are the downsides to just always setting strict? I think it's "just" that if a schema is provided the OpenAI doesn't understand, it'll fail? Should we just always set strict and then optionally use something in additional properties as an override?

What are the downsides to just always setting strict? I think it's "just" that if a schema is provided the OpenAI doesn't understand, it'll fail? Should we just always set strict and then optionally use something in additional properties as an override?

There's nothing in OpenAI's announcement that clarifies the drawbacks. Maybe they just didn't know whether it would always work.

As per this longer explanation, I'd support always setting strict if there's a schema. And like you say, if OpenAI really needs to support some non-strict mode, that's very OpenAI-specific and could be handled by something in additional properties to override this.

I pushed an update to your branch to fix up the remaining strict handling.