구조화된 출력 퀴즈

퀴즈를 풀면 정답과 설명이 나타납니다.

1. A JSON Schema declares a property named total. What must also be specified if every response must include it?
   • Set its numeric type only
   • Add total to the required-field list
   • Place it first in the schema file
   • Give the object a longer description
2. Why decide whether additional object fields are allowed?
   • Any extra property always proves hallucination
   • JSON syntax forbids all additional properties
   • Only arrays can contain unexpected fields
   • Unexpected fields can change how downstream code interprets a response
3. A missing value and an explicit null have different meanings in an integration. What should the schema specify?
   • Their separate allowed behavior and downstream interpretation
   • Treat absence and null as the same without documentation
   • Replace both with a fabricated zero
   • Let every provider choose silently
4. A receipt has subtotal 50 and tax 5; valid JSON reports total 550. What failed?
   • JSON parsing
   • The total field’s numeric type
   • Factual and arithmetic validation against the source
   • The existence of an object schema
5. A structured result cites source_id 9, but only IDs 1 through 3 were supplied. What check catches the error?
   • Check that source_id is a string
   • Count the object’s top-level keys
   • Accept any integer because its type is valid
   • Verify that referenced identifiers actually exist in the supplied evidence
6. How should an application handle a model refusal or truncated response?
   • Assume every schema field was produced
   • Treat it as an explicit non-success case rather than inventing required values
   • Silently fill missing evidence with plausible defaults
   • Ignore the boundary because JSON mode was requested
7. A stored output uses schema v1 but a consumer expects v2. What should the integration do?
   • Apply v2 assumptions without checking
   • Remove the version field
   • Reject or explicitly adapt the incompatible version
   • Assume schema changes affect only formatting
8. Two providers offer structured output, one constrains generation and one validates afterward. Which practice remains necessary for both?
   • Check supported schema features and verify the values’ meaning
   • Assume identical enforcement for every schema keyword
   • Skip factual checks when output parses
   • Treat post-validation as a guarantee of truth