CCAR-P sample question: when a stakeholder wants the same answer every time

An original sample question by CERTain·exam, based only on Anthropic's public documentation and the public exam guide. CERTain·exam is not affiliated with Anthropic.

Domain 6 · Stakeholder Communication & Lifecycle Management (14% of CCAR-P in the exam guide effective until December 6, 2026; 7% from December 7, 2026). Objective covered: "Manage stakeholder feedback loops and expectation alignment (including SLAs)."

This is a multiple-response question with two correct answers. It tests whether you separate what a stakeholder asks for from what she actually needs, and whether you correct an expectation the platform cannot meet. Work through it before you reveal the answer.

The question

Wrenfield Building Society's members ask a Claude Sonnet 5.5 assistant about savings accounts. Requests now use API version 2026-10-21. When the same question was asked twice, it got two differently worded answers, both correct. The head of member services wants the team to "switch the randomness off" as in last year's pilot, which set temperature to 0 on an older model. She worries that members who compare answers will think the terms differ, so the rate, term, and withdrawal rules must always look the same.

Which TWO actions should the architect take? (Select TWO.)

  • A. Pin the assistant to the older API version so that temperature 0 works again, and promise members identical replies.
  • B. Tell her no current setting makes replies repeat word for word; the same request can get differently worded replies.
  • C. Put rate, term and withdrawal rules in fixed output fields, leaving Claude's own wording to a short note.
  • D. Add a system-prompt instruction to answer a repeated question in exactly the same words as the last time.
  • E. Generate three answers per question and send one only if all three match word for word; otherwise, pass the question to staff.
Show the answer and explanation

Answer

B. Tell her no current setting makes replies repeat word for word; the same request can get differently worded replies.

C. Put rate, term and withdrawal rules in fixed output fields, leaving Claude's own wording to a short note.

How to read this question

  1. Situation. A stakeholder asks to "switch the randomness off" as in an older pilot. The current model accepts only default sampling, and her real need is that rates and terms always appear the same way.
  2. What is asked. Two actions: what the architect must tell her about non-determinism, and which design change delivers the consistency she actually needs.
  3. Option by option.
    • A. Claude 4.7 and later models accept only the default sampling values, so an older API version brings no temperature control to Sonnet 5.5. And even at temperature 0, identical inputs can produce different outputs.
    • B (correct). The architect corrects the expectation with a platform fact. The sampling settings are gone, nothing replaces them, and the same request can still get differently worded replies. Word-for-word repetition is not something to promise.
    • C (correct). Her real need is that the facts look the same every time. A fixed output format puts rate, term and withdrawal rules in the same fields on every answer, so variation is confined to a short note.
    • D. The model cannot reproduce a previous reply on request. The same input can produce different outputs, so the instruction promises what generation cannot guarantee. Each call also has no record of the replies given earlier (reasoning from the scenario).
    • E. Since wording varies between identical requests, three-way word-for-word agreement will rarely occur, so almost every question would go to staff, at triple the cost (reasoning from the glossary statement).
  4. Rule to remember. Generated wording varies even for the same input, and current models offer no setting to stop it. Tell stakeholders so, and make consistency a property of the output structure, not of the sampling.

Sources (Claude Docs, quotes checked Oct 28, 2026)

  • Glossary, "Temperature":
    • "Even with temperature set to 0, the results will not be fully deterministic, and identical inputs might produce different outputs across API calls."
  • Migrating to API version 2026-10-21, "Sampling parameters are removed":
    • "Models that already accept only the defaults, such as Claude 4.7 and later models, sample as before."
    • "No other setting replaces it." (said of requests that lowered temperature on models that accepted other values, such as Claude Sonnet 4.6)
  • Increase output consistency:
    • "Precisely define the output format you want using JSON, XML, or custom templates so that Claude follows every output formatting element you require."

More practice

Questions

Where does this question come from?

It's an original sample question that CERTain·exam wrote based on Anthropic's public documentation and the public exam guide. Exam content is confidential under Anthropic's candidate agreement, and we do not use it.

What does CCAR-P Domain 6 cover?

According to the exam guide effective until December 6, 2026, Stakeholder Communication & Lifecycle Management (14%) covers structured discovery and requirement gathering, communicating architectural decisions and trade-offs, managing stakeholder feedback loops and expectation alignment (including SLAs), documenting architectures and implementation guidance, and supporting lifecycle phases from discovery to iteration. From December 7, 2026, the domain weighs 7%, and its objectives are rewritten. See what changes on December 7, 2026.

Get your free practice exam

First, sign in with your email.