Use decision models for structured questions

Use decision models to answer a defined set of questions about text or JSON, rather than asking a general chat model for free-form output. Discover supported models, publish a question-set version, and use that version for repeatable judgments.

The API, SDK, and MCP expose model discovery and question-set execution. Results retain per-question refusals and provider confidence where available, so a refusal need not be mistaken for a confident answer.

This release covers developer access. The Question sets dashboard is still under a separate rollout flag, and each provider's credential and retention requirements still apply.