Distinguish response formatting from actions
Structured output constrains a response format; tool calling expresses requests to invoke application-defined capabilities. Your application still validates arguments, enforces permissions and handles errors. A supported-parameters list does not prove that a model will always produce valid results.
Check the selected endpoint
Model-level capability claims can differ from provider endpoint support. Review the offer and source documentation for the parameters you need. Use the same endpoint, configuration and schema in tests that you intend to use in production.
Test failure handling
Include malformed inputs, missing information and unsupported operations in your evaluation. Measure schema adherence and task correctness separately. The exchange lists declared capabilities and links to providers; it does not execute or authorize application actions.