Structured output and validation

Methods to understand

Structured output requires a schema and server-side validation. Check required fields, types, enums, and lengths. Reject missing fields instead of guessing business values from unparseable text.

Apply the method to a scenario

Goal: complete a clearly scoped work task using AI.
Input: public, task-relevant material.
Acceptance: verifiable results and explicitly marked unknowns.

Hands-on workshop

Independently validate fields in model output.

1. Define required fields, types, and limits.
2. Validate parsed JSON before passing it to downstream business logic.
3. Construct missing-field, wrong-type, and unknown-field cases.

function validateAnswer(value) {
  if (!value || typeof value !== "object") throw new Error("Invalid answer");
  if (typeof value.answer !== "string" || value.answer.length > 2000) throw new Error("Invalid text");
  if (!Array.isArray(value.sources) || value.sources.some(s => typeof s !== "string")) throw new Error("Invalid sources");
  return { answer: value.answer, sources: value.sources };
}

How to verify

Also check that source identifiers exist in actual retrieval results. A correct structure does not guarantee correct content.

Practice task

Choose a case related to your work and practice "Structured output and validation". Submit inputs, steps, results, and verification records.

• Explain the task goal and inputs
• Provide actual results or runnable deliverables
• Identify errors, limits, and verification evidence