NCA-GENL Software Development Practice Question
A team is building a RAG assistant and wants to reduce hallucinated citations. They plan to have the LLM return structured output that names the source document chunk used for each claim. Which implementation strategy most directly improves the reliability of that structured output?
⚠ Common exam trap
The trap here is assuming that asking the model politely for citations produces trustworthy citations, when only constrained decoding against known identifiers makes them verifiable.
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Pass the retrieved chunks with stable identifiers in the prompt and constrain generation to a JSON schema that references those identifiers.
Reliable attribution comes from making the citation a constrained choice among known identifiers. Supplying stable chunk IDs in context and enforcing a JSON schema means the model selects from real evidence instead of generating citation text. The application can then validate each returned identifier against the retrieved set, converting an unverifiable prose claim into a programmatically checkable reference.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Retrieve more chunks and place them all in the prompt without identifiers.
Why it's wrong here
Adding unlabeled chunks increases context length and cost while making it harder for the model to indicate which passage supported a claim. Without identifiers there is nothing for the model to reference precisely, so citations remain vague or invented. More context alone does not create the verifiable link between claim and source that structured identifiers provide.
- ✗
Increase the model temperature so the model explores more citation candidates.
Why it's wrong here
Higher temperature increases randomness in token selection, which makes structured output less consistent and increases the chance of malformed or invented citations. For a task that depends on faithful attribution, deterministic decoding is preferable. Raising temperature works against the goal of reliable, repeatable structured output and can degrade adherence to the requested schema.
- ✗
Ask the model to cite sources in a free-text bibliography appended after the answer.
Why it's wrong here
Free-text bibliographies are not constrained to the retrieved set, so the model can produce plausible but nonexistent titles or authors that are hard to validate programmatically. Parsing them reliably is also fragile. Structured identifiers tied to the actual retrieved chunks allow deterministic verification, whereas prose citations leave hallucination detection to subjective review.
- ✓
Pass the retrieved chunks with stable identifiers in the prompt and constrain generation to a JSON schema that references those identifiers.
Why this is correct
Giving the model explicit, stable chunk identifiers in context makes attribution a selection task rather than an invention task. Constraining decoding to a JSON schema that requires an identifier field prevents free-form citation text and makes the output machine-checkable. The application can then verify that every cited identifier exists in the retrieved set, catching unsupported claims before they reach the user.
About these practice questions
This NCA-GENL question is part of Courseiva's 367-question bank — original exam-style content with full explanations and wrong-answer analysis, never real exam questions or exam dumps. Learn why practice questions differ from exam dumps →
JA
Written and reviewed by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
Last reviewed September 2026 · checked against the official NVIDIA exam blueprint
This NCA-GENL practice question is part of Courseiva's free NVIDIA certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the NCA-GENL exam.