Architecture scenarioMedium~10 minPrompting & Structured Output / 16 of 20

Interview question

How should an application handle refusals, incomplete output, and truncation?

Separates provider refusals, safety blocks, token truncation, transport failure, and valid domain abstention.

1Say itAnswer out loud first

2CompareCompare with Core answer

3RefineRefine with Stronger and Advanced

TL;DR

I inspect the provider’s response status before parsing the success schema. A refusal or safety block, token-limit truncation, timeout, cancelled stream, and valid insufficient-evidence result are different outcomes. I map each one to an explicit application state and a useful user response.

Full answer below

Practice focus

Whether distinct model and transport outcomes receive distinct application behavior.

Answer first

Answer path

Core answer first, then checkpoints toward a stronger answer.

Start
Core answer
Start here. This is the direct spoken answer to practice first.
Overview
A missing success payload is not one generic parsing error. The reason determines whether to retry, ask for different input, fall back, or stop.
I inspect the provider’s response status before parsing the success schema. A refusal or safety block, token-limit truncation, timeout, cancelled stream, and valid insufficient-evidence result are different outcomes. I map each one to an explicit application state and a useful user response.
Checkpoint
Stronger answer
Optional upgrade with examples, trade-offs, and sharper phrasing.
Truncation may justify reducing context, increasing an allowed output budget, or continuing only if the protocol supports safe continuation. A refusal normally should not be blindly retried with the same request. A network timeout may be retryable, but only within a deadline and with idempotent downstream behavior. I keep the original failure classification in telemetry.
Finish
Advanced answer
Optional deeper angle for production judgment and nuance.
I map provider-specific stop reasons into stable internal categories while retaining the raw diagnostic code for support. A refusal is therefore not confused with a truncated response or a schema failure. Metrics separately track policy refusal, capacity or rate limits, malformed output, semantic rejection, and user cancellation. A fallback model is allowed only when its policy, schema, and data-handling requirements remain compatible with the original request.

Review signals

Use these to catch weak phrasing, shallow reasoning, and risky claims.

Common mistakesCandidateCommon mistakesMistakes that weaken the answer or signal shallow understanding.Open details

Details

Parsing every non-success response as malformed JSON.
Retrying a refusal repeatedly with identical input.
Showing a partial structured object as a complete result.
Hiding provider stop reasons behind one generic error.

Red flagsCandidateRed flagsSignals that an interviewer may treat as risky.Open details

Details

Truncated output can trigger a partial side effect.
Fallback silently weakens policy or data boundaries.
The system cannot tell refusal from dependency failure.

Follow-up prompts

Quick interviewer prompts inside this drill.

Follow-up questionsCandidateFollow-up questionsLikely next questions after the main answer.Open details

Details

Treat these as the next interviewer turn. Answer one out loud, then save the gap in your practice notes.

Which outcomes are retryable?
When is continuation safe?
How would you map provider-specific stop reasons?

Continue the section

Next in Prompting & Structured Output: Repair invalid output

Step 17 of 20

How would you repair invalid structured model output?

Practice route

Read core answer Review checklist Follow-up prompts Track progress

Finished this drill?

Practice freely. Sign in only when you want to save completion, confidence, notes, or bookmarks.

New to Aporeon? Create account

Drill details

Human editorial direction

Aporeon is shaped by Aleksandar Tomovski, a software developer with experience on both sides of technical interviews. Content is reviewed for accuracy, natural spoken delivery, useful depth, and honest trade-offs.

About the creator Editorial standards

Was this drill useful?

Help improve the interview library.

Found this answer useful?

Say thanks with a standalone, one-time $5 contribution. No account required.

No subscription or perks.Choose another amount

Opening drill

Architecture scenarioMedium~10 minPrompting & Structured Output / 16 of 20

Interview question

How should an application handle refusals, incomplete output, and truncation?

Separates provider refusals, safety blocks, token truncation, transport failure, and valid domain abstention.

1Say itAnswer out loud first

2CompareCompare with Core answer

3RefineRefine with Stronger and Advanced

TL;DR

Full answer below

Practice focus

Whether distinct model and transport outcomes receive distinct application behavior.

Answer first

Answer path

Core answer first, then checkpoints toward a stronger answer.

Start
Core answer
Start here. This is the direct spoken answer to practice first.
Overview
A missing success payload is not one generic parsing error. The reason determines whether to retry, ask for different input, fall back, or stop.
I inspect the provider’s response status before parsing the success schema. A refusal or safety block, token-limit truncation, timeout, cancelled stream, and valid insufficient-evidence result are different outcomes. I map each one to an explicit application state and a useful user response.
Checkpoint
Stronger answer
Optional upgrade with examples, trade-offs, and sharper phrasing.
Truncation may justify reducing context, increasing an allowed output budget, or continuing only if the protocol supports safe continuation. A refusal normally should not be blindly retried with the same request. A network timeout may be retryable, but only within a deadline and with idempotent downstream behavior. I keep the original failure classification in telemetry.
Finish
Advanced answer
Optional deeper angle for production judgment and nuance.
I map provider-specific stop reasons into stable internal categories while retaining the raw diagnostic code for support. A refusal is therefore not confused with a truncated response or a schema failure. Metrics separately track policy refusal, capacity or rate limits, malformed output, semantic rejection, and user cancellation. A fallback model is allowed only when its policy, schema, and data-handling requirements remain compatible with the original request.

Review signals

Use these to catch weak phrasing, shallow reasoning, and risky claims.

Common mistakesCandidateCommon mistakesMistakes that weaken the answer or signal shallow understanding.Open details

Details

Parsing every non-success response as malformed JSON.
Retrying a refusal repeatedly with identical input.
Showing a partial structured object as a complete result.
Hiding provider stop reasons behind one generic error.

Red flagsCandidateRed flagsSignals that an interviewer may treat as risky.Open details

Details

Truncated output can trigger a partial side effect.
Fallback silently weakens policy or data boundaries.
The system cannot tell refusal from dependency failure.

Follow-up prompts

Quick interviewer prompts inside this drill.

Follow-up questionsCandidateFollow-up questionsLikely next questions after the main answer.Open details

Details

Treat these as the next interviewer turn. Answer one out loud, then save the gap in your practice notes.

Which outcomes are retryable?
When is continuation safe?
How would you map provider-specific stop reasons?

Continue the section

Next in Prompting & Structured Output: Repair invalid output

Step 17 of 20

How would you repair invalid structured model output?

How should an application handle refusals, incomplete output, and truncation?

Practice focus

Answer path

Overview

Review signals

Follow-up prompts

Continue the section

How should an application handle refusals, incomplete output, and truncation?

Practice focus

Answer path

Overview

Review signals

Follow-up prompts

Continue the section