The useful result
Choose a safe next step for each response state before validating and using its data.
- Choose a safe next step for each response state before validating and using its data.
Schema validation starts too late if your application mistakes a refusal or interrupted response for a normal empty result. OpenAI documents separate handling for these states. Practise a small decision table before connecting an action to model output. The fixtures here are normalized application examples, not a claim about an SDK’s exact raw response shape.
Select the named model only if it is available in your interface. API controls and consumer-app controls differ. You can practise the technique on the supplied fictional inputs with an available compatible assistant first. This lesson is documentation-based; examples are expected results, not a claim that we ran the named model.
Before you start
- A test environment or paper decision table; no live API call required.
- Four normalized fixtures: refusal, incomplete, completed missing fact, completed valid record.
Review evidence
Documentation-based review · 2026-10-18
Provider documentation checked on 18 October 2026. Examples and checks are original illustrative exercises, not recorded platform tests. Cover artwork is conceptual and does not imply provider sponsorship.
Separate transport state from business facts
Write the order of checks: request/error state, completion state, refusal content, parsed payload and schema validity, then factual review. Follow your API’s actual response documentation when implementing. Do not use null alone to represent every failure.
Example / expected result
Illustrative example; not a recorded model result.
Normalized cases: A refused; B incomplete; C completed with due_date=null; D completed with a source-supported due_date.
Check your result
Your design can distinguish a missing fact from a response that never completed.
Choose a distinct action for each case
A refusal needs an appropriate explanation or alternative; an incomplete result needs controlled recovery; a valid missing value stays unknown; a complete value still needs source checking. These decisions are outside the model’s prose.
Review this application decision table without calling any API: refused → explain; incomplete → stop and recover; completed with due_date=null → retain unknown; completed with a supplied date → validate and check source. Identify any branch that wrongly treats refusal as a normal empty record.Example / expected result
Illustrative example; not a recorded model result.
A: no downstream action. B: retain partial output privately, recover under a bounded policy. C: unknown date, request source if required. D: validate and compare to supplied evidence.
Check your result
No refused or incomplete fixture reaches the ordinary record-action path.
Test every branch before deployment
Use mocked responses in your own application and assert which branch executes. Include a malformed payload and a schema-valid but unsupported date. Keep retry limits explicit; a repair loop must not endlessly resubmit a refusal or silently invent a value.
Check your result
Tests demonstrate separate branches, and factual checks still reject an unsupported date.
When it goes wrong
Structured output constrains shape, not truth. Exact response envelopes and SDK helpers vary; the normalized fixtures are not copy-ready SDK code. Do not defeat a refusal by automatic retries or treat a network error as a valid “no data” answer.
Make it your own
Add a completed schema-valid date that does not appear in the source. It should pass the shape check and fail the separate evidence check.
Useful next resources
Sources and review
Last checked Oct 18, 2026. Features and account limits can change. Verify important results against your original material.
Did this help you complete your task?
Tell us what needs improving. No account or private output required.
Report incorrect information or a broken linkGenius Forges Editorial
Practical AI lessons with checked sources, labelled examples and human verification.
