Why Are Function Calls Executed When Required Properties Are Not Inferred?

Below is an explanation of how the required property validation works in the context of function calling:

When Does Validation Happen?

The validation against your JSON Schema (which includes the required properties) is performed as part of the post-generation output cleaning process. In effect, the model generates an output and then that output is validated and transformed according to your schema. This means that if a required property is missing or empty, the error isn’t necessarily caught before the function call is attempted. Instead, it is caught during the subsequent validation step.

What Does This Mean for Your Use Case?

• If some required properties are empty, the function call may still be fired, and the function will receive those empty values.
• The strict validation (for example, by setting strict: "true", or using additional validation libraries like Pydantic) occurs after generation. It ensures later on that the data adheres to the defined schema.
• The model does not automatically trigger a fallback to regular chat if required properties are missing. Instead, it packages the generated output (even if it’s missing a required property) into the function call.

How Can You Handle Missing Required Properties?

1. Pre-Validation in Your Code:
   Implement your own pre-checks in your application code. If a required property is empty, you can choose to skip the function call or trigger a follow-up request (e.g., asking the user for clarification or additional details).

2. Structured Outputs:
   Ensure you have structured output enabled (using strict: "true"). This guarantees that your model’s output aligns closely with your schema, but remember that the schema validation is applied after generation.

3. Output Cleaning and Post-Processing:
   After the model generates the output, run your transformation and error-handling routines. If required properties are missing, have your system either handle them gracefully (by supplying defaults or asking for more input) before proceeding with the function call.

Conclusion

The required property validation is applied post-generation; it doesn't prevent the function call from being made if required items are missing. To ensure chat continues as normal when some required properties are empty, you need to implement additional logic in your application. This logic can detect improperly filled properties and decide whether to prompt the user for more information or handle the situation differently.

By handling these validations on your side—either before sending the function call or right after—you can ensure your application behaves as expected even if some required inputs are missing.