CCAR-P Advanced Agentic Architecture Practice Question
An agent orchestrator delegates work to three specialist sub-agents: a 'search' agent that returns ranked documents, an 'extract' agent that pulls structured fields from those documents, and a 'verify' agent that checks extracted fields against source text. During evaluation you find that verify frequently approves fields that extract hallucinated, because verify receives only the extracted JSON, not the source passages. Which change to the orchestration contract most directly fixes this?
⚠ Common exam trap
The trap here is treating verification failures as a reasoning or sampling problem and adding temperature, re-ranking, or a redundant extractor, when the verifier simply lacks the source evidence needed to detect fabrication.
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
✓
Have the extract agent include, for each field, the source span it was derived from, and pass those spans to the verify agent alongside the extracted JSON.
The verifier approves hallucinations because it is asked to judge extracted values without access to the text they were derived from. Propagating per-field source spans gives the verifier the evidence needed to confirm or refute each value against its origin. This is a contract change between extract and verify, ensuring provenance flows with the data rather than being reconstructed or guessed downstream.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Route all extracted fields back through the search agent so it can re-rank the documents before verification.
Why it's wrong here
Re-ranking changes which documents are presented, not whether the verifier can see the text behind a field. The search agent has no knowledge of which span the extract agent used, so it cannot supply the missing evidence. This adds a hop and latency without giving the verifier the source material it needs to judge a specific extracted value.
- ✗
Increase the verify agent's temperature so it is more likely to question the extracted fields it receives.
Why it's wrong here
Temperature changes sampling diversity, not the information available to the verifier. A verifier that never sees the source passages has no basis for detecting a hallucinated field regardless of how creatively it samples. Raising temperature may make verification less consistent without improving its ability to catch fabrication, so it does not address the missing-evidence root cause.
- ✗
Add a fourth 'audit' agent that independently re-runs the extract agent on the same documents and compares the two JSON outputs.
Why it's wrong here
Comparing two extraction passes detects instability but not fabrication: both passes can confidently produce the same hallucinated value if the source lacks the field and the model fills it plausibly. It also doubles extraction cost. Without grounding either pass in source spans, the audit cannot distinguish a correct value from a consistently invented one.
- ✓
Have the extract agent include, for each field, the source span it was derived from, and pass those spans to the verify agent alongside the extracted JSON.
Why this is correct
Verification is only possible against evidence. By propagating the exact source spans that back each extracted field, the verify agent can compare the claimed value to the text it supposedly came from, which is the only way to detect fabrication. This changes the orchestration contract to carry provenance, directly closing the gap that let hallucinated fields pass.
About these practice questions
One of 262 original CCAR-P practice questions on Courseiva, each with a full explanation and wrong-answer analysis — not exam dumps or protected exam content. 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 Anthropic exam blueprint
This CCAR-P practice question is part of Courseiva's free Anthropic 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 CCAR-P exam.