CCAR-P domain - 14% of the exam

Stakeholder Communication & Lifecycle Management

Stakeholder Communication & Lifecycle Management is 14% of the CCAR-P exam. These are the objectives it covers, each with practice questions, with every answer explained.

The domain in numbers

45
Practice questions
5
Objectives
10
Easy
35
Medium

Where people go wrong

  • Fine-tuning before prompting and retrieval

    The wrong answers propose fine-tuning on two years of approved replies, 600 inconsistent labelled decisions, past support replies or last term's transcripts before prompting and retrieval have been tried. Fine-tuning bakes knowledge into weights, so it goes stale when policy or a syllabus is revised and cannot cite which clause version it used. The correct answers retrieve current material at request time, such as a versioned policy index or the revised syllabus, and keep labelled data for evaluation instead of training.

    See it in a sample question
  • Expecting temperature, caching or voting to guarantee identical output

    Lowering sampling temperature to zero, adding a response cache or taking a majority vote each looks like a way to guarantee identical output. None does: output can differ between runs even at temperature zero, a cache applies only to byte-identical input, and a vote is itself probabilistic. Where two identical cases must get the same outcome, the correct answer moves the decision into deterministic rules code. Where identity cannot be promised, it records each reply's inputs so the reply can be explained later.

    See it in a sample question
  • Giving a review board reassurance in place of the facts it decides on

    The wrong answers hand the reviewer something easier than what it asked for: the two-day benefit alone, a monitoring plan, an encryption statement, or a prompt rule presented as making broad write access negligible risk. A security board accepts residual risk and a privacy officer asks, place by place, what is held, why and for how long. The correct answers state those facts directly: the exact write actions and records in reach with their scoping controls, or a data flow record giving each hop's purpose, retention period and deletion mechanism.

    See it in a sample question

Objectives in this domain

What to study

The official documents this domain's practice questions cite most.

Sample questions from this domain

Free sampleStakeholder Communication & Lifecycle Managementmedium

A retail bank wants Claude to draft replies to customer complaints. Discovery has established that the team handles about 4,000 complaints a month, that the complaints policy library holds 2,300 documents which the compliance team revises weekly, and that the regulator requires each reply to cite the policy clause, and its version, that was in force on the date of the complaint. The sponsor proposes fine-tuning a model on two years of approved replies. What should the architect recommend?

  • AFine-tune a model on the two years of approved replies so policy positions and house style are carried in the model weights.
  • BLoad the whole policy library into the context window on each request so that retrieval cannot miss a relevant policy clause.
  • CPrompt the model with a summary of current policy positions, and have a reviewer add the clause citations before each reply is sent.
  • DRetrieve clauses from a versioned policy index filtered by complaint date, and have each draft cite the retrieved clause and version. Correct
When discovery reveals frequently revised source material and a version-specific citation duty, the architecture should retrieve from a versioned index rather than fine-tune. Discovery answers about data freshness and traceability decide the architecture. Fine-tuning bakes knowledge into weights that go stale on every revision and cannot attribute an answer to a source. Retrieval keeps knowledge outside the model, so a weekly revision is live once indexed, a date filter selects the version in force on the complaint date, and the retrieved text is what the reply cites to satisfy the regulator.

Why A is wrong: This is tempting because the sponsor proposed it and two years of approved replies look like ideal training data for tone and policy. It is wrong because a weekly revision cycle would leave the weights stale within days, and a fine-tuned model cannot point to which clause version it relied on, so the citation requirement cannot be met.

Why B is wrong: This is tempting because it removes retrieval misses as a failure mode. It is wrong because 2,300 documents on each of 4,000 requests a month is costly and slow, the model still has to pick the version in force on the complaint date from a mass of competing versions, and long-context stuffing is the over-engineered answer where targeted retrieval fits.

Why C is wrong: This is tempting because it keeps a human in the loop on a regulated output and avoids building an index. It is wrong because a summary of current positions cannot reflect the version in force on an older complaint date, and moving the citation duty to a manual step is a compensating control that adds cost to every reply where a structural fix exists.

Why D is correct: Correct. The discovered facts (weekly revisions, a large library, and a date-specific citation duty) all point to retrieval over a versioned index: updates reach answers as soon as they are indexed, the date filter selects the version in force, and the retrieved clause gives the draft something concrete to cite.

Free sampleStakeholder Communication & Lifecycle Managementmedium

A government benefits agency uses Claude to read applicants' supporting documents and decide eligibility for a hardship grant, about 4,000 applications a week. An audit finds that resubmitting the same application can produce a different eligibility outcome in a small share of cases. Policy requires that two applicants with identical eligibility facts receive the same determination, and the agency cannot add caseworker headcount. Which change should the architect recommend to the programme board?

  • AHave Claude extract the eligibility facts into a fixed schema and let deterministic rules code make the determination from those facts Correct
  • BSet the sampling temperature to zero and report to the board that this makes the model's eligibility decisions fully reproducible from now on
  • CRoute every eligibility decision to a caseworker for confirmation before the outcome is issued, so that a person resolves any inconsistency
  • DRun each application through the model three times and issue whichever eligibility outcome appears in the majority of the three runs
Where policy demands identical outcomes for identical facts, use the model for extraction and make the decision in deterministic code rather than promising model reproducibility. Model output is probabilistic even with conservative sampling settings, so no prompt or sampling change can promise identical determinations. Separating extraction from decision means the varied, unstructured documents are handled by the model while the eligibility outcome comes from rules code that is deterministic for a given set of facts. The expectation set with the board then matches what the architecture can guarantee, and the volume constraint is respected.

Why A is correct: Splitting the task puts the probabilistic step where it is useful, reading varied documents, and moves the decision into rules code that returns the same outcome for the same facts. That meets the consistency policy structurally, at the existing volume, without extra caseworkers.

Why B is wrong: Lowering temperature is tempting because it does reduce variation between runs. It is wrong because it does not guarantee identical outputs, so the board would be given an expectation the system cannot meet and the policy requirement would still be breached occasionally.

Why C is wrong: Human confirmation is tempting because a person can spot inconsistent outcomes. It is wrong because the agency cannot add headcount and 4,000 weekly reviews would overwhelm existing staff, and it places review on every case rather than fixing the source of the inconsistency.

Why D is wrong: Majority voting is tempting because it does reduce the effect of a single odd run. It is wrong because the vote itself is still probabilistic, so identical applications can still diverge, and it triples processing cost without meeting the policy requirement structurally.

Free sampleStakeholder Communication & Lifecycle Managementeasy

A US telehealth provider plans to launch a Claude-powered assistant that triages patient messages before a nurse reviews them. Patient text passes through the web application, a request log, a vector store of past cases, the model API and an analytics warehouse. The privacy officer will not approve go-live until the provider can show a regulator, for each place patient data goes, what is held there, why, and when it is deleted. The audit is in four weeks. What should the architect produce?

  • AA system prompt instruction telling the assistant not to retain or repeat patient identifiers beyond the conversation
  • BA security statement confirming that each data store is encrypted at rest and in transit with managed keys
  • CA data flow record listing each hop, the data it holds, the purpose, the retention period and the deletion mechanism Correct
  • DA change switching off the request log and the analytics warehouse so that patient data is held in fewer places
A data flow and retention record, listing each place data goes, its purpose and its deletion period, is the handoff artefact that answers privacy and regulatory review. Regulators and privacy officers assess an LLM system by where personal data travels, not by how the model is prompted. Secondary stores such as request logs, embeddings in a vector store and analytics copies often hold the same patient text as the primary application. Recording each hop with its purpose, retention period and deletion mechanism gives the evidence needed for review and makes data minimisation decisions visible, which encryption statements or prompt instructions cannot do.

Why A is wrong: Tempting because it appears to minimise patient data. It is wrong because an instruction to the model controls nothing about what the application, log, vector store or warehouse keep, and an instruction is not evidence a regulator can audit.

Why B is wrong: Tempting because encryption is a real and expected control for patient data. It is wrong because encryption protects data where it sits; it says nothing about which places hold patient data, for what purpose, or when it is deleted.

Why C is correct: Correct. The privacy officer's question is asked per location: what is held, why, and for how long. A data flow and retention record answers it hop by hop, including the vector store and warehouse that are easy to overlook, and shows the mechanism that enforces each deletion period.

Why D is wrong: Tempting because fewer stores means less to account for. It is wrong because it removes logging needed to investigate incidents and analytics the provider relies on, and the remaining stores still need documenting, so it trades away function without answering the audit.

Other domains in this exam

See also the CCAR-P cert hub, the study guide, and the cheat sheet.

Examworthy is not affiliated with or endorsed by Anthropic. Original, blueprint-aligned practice material only.