CCAR-P - Stakeholder Communication & Lifecycle Management (14% of the exam) - Section 6.5

Support lifecycle phases (discovery, design, handoff, monitoring, iteration).

Carrying a solution through discovery, design, handoff to the owning team, production monitoring and iteration. Candidates should recognise the risks at each transition, especially a handoff with no owner for ongoing evaluation.

discoverydesignhandoff and ownershipmonitoring and iteration

Practice question for this objective

Free sampleStakeholder Communication & Lifecycle Managementmedium

A national benefits agency's caseworker summarisation assistant has run in production for four months, built and operated by an external vendor that runs a weekly evaluation against 500 labelled case files. The contract ends next month and the service passes to the agency's in-house platform team, which has no machine learning specialists but already runs the agency's service dashboards. Agency policy requires the agency itself to detect any drop in summary quality within one week. What should the architect require before the vendor exits?

AHave the vendor deliver a final evaluation report, a written runbook and the prompt history, so the in-house team understands how the summariser was built and tuned
BKeep the vendor on a reduced support contract to investigate and fix any summary quality problems that caseworkers raise through the agency's service desk
CTransfer the evaluation suite and labelled set, name an in-house owner for it, and schedule it to run weekly with an alert when scores fall below the current baseline Correct
DAdd latency, error-rate and token-usage alerts to the in-house team's existing dashboards, so degradation in the summariser is visible on the screens they watch

A production handoff must transfer ongoing evaluation with a named owner, not just documentation, or quality regressions go undetected once the builder leaves. The most common failure at handoff is that the evaluation stops because it was the builder's activity rather than part of the service. Moving the labelled set and evaluation into an automated weekly job, with a named in-house owner and an alert against the current baseline, keeps quality measured without needing specialist skills to run it. Documentation, operational dashboards and reactive vendor support each leave the agency unable to detect a quality drop itself within a week.

Why A is wrong: This is tempting because documentation is a normal part of any handoff and helps the new team understand the system. It is wrong because a final report is a snapshot: once the vendor leaves, nobody runs the weekly evaluation, so a quality drop would go undetected and the one-week detection policy fails.

Why B is wrong: This is tempting because it retains the expertise that built the system. It is wrong because it is reactive: quality problems are found only when caseworkers notice and report them, which breaks the requirement that the agency itself detect a drop within a week.

Why C is correct: This is correct because it keeps the quality measurement running after the handoff and gives it a named owner, and an automated scheduled run with a baseline alert does not require machine learning specialists to operate, so the one-week detection requirement continues to be met.

Why D is wrong: This is tempting because it fits the in-house team's existing skills and tooling. It is wrong because those are operational health signals; a summariser can stay fast, error-free and on budget while its summaries become less accurate, so this does not measure quality at all.

See more CCAR-P practice questions, answers explained.

Exam traps in Stakeholder Communication & Lifecycle Management

Answers that look right on this material and are not. Each one is a distractor from a different question in the CCAR-P bank for this domain.

A hard spending cap that rejects teacher requests once the monthly budget is reached, so the department cannot overspend
Why it is wrong: Tempting because it guarantees the budget is never exceeded. It is wrong because the cap is most likely to trip in the exam weeks, when usage triples, and rejecting requests then breaks the stated requirement that the assistant stay available to teachers through that period.
The provider changed the model weights after the evaluation, degrading routing of account-access tickets
Why it is wrong: Tempting because blaming the model is a common first reaction to a production drop. It is wrong because the assistant is still routing exactly as the runbook says; its behaviour is consistent, and the mismatch lies in the rules it was given.
Thumbs-down feedback is too noisy to act on, so the stable evaluation score is the more reliable quality signal.
Why it is wrong: User feedback is noisy and an offline score is controlled, so preferring the score feels disciplined. It is wrong here because the thumbs-down rate nearly tripled over ten weeks in step with a known shift in question mix, which is a sustained signal, not noise.

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