Part 9f — Question Bank: Domain 6, Stakeholder Communication & Lifecycle Management (Q120–Q140)
Original practice questions; not official or leaked. Objectives paraphrase the D6 task statements in Exam Guide v1.0 (see 00_Exam_Scope_and_Study_Plan.md). Multi-response items (Q127, Q137) say "(Select TWO)" and have 5 choices.
Q120 · D6 · Objective: Elicit and clarify business requirements through discovery · Difficulty: Intermediate
Scenario: A VP says, "We need an AI agent for customer support." No metrics, volumes, or constraints have been shared. Question: What should the architect do first? A. Start building a prototype with the most capable model. B. Run discovery: the target workflow, baseline volumes and costs, success metrics, data sources, risk tolerance, and constraints, then confirm the problem is a fit for an LLM solution. C. Select a vector database. D. Write the system prompt. Correct answer: B Explanation: Without a defined problem and measurable success criteria, every downstream choice is a guess. Why not the others: A, C, and D are solution steps taken before requirements exist. Architectural principle: Problem and metrics before solution. Ref: Prep Lesson 4
Q121 · D6 · Objective: Elicit and clarify business requirements through discovery · Difficulty: Advanced
Scenario: Two executives give conflicting goals: Operations wants minimum cost per ticket; Legal wants zero unreviewed customer-facing statements. Question: What is the best approach? A. Pick the more senior executive's goal silently. B. Build to cost and handle Legal later. C. Surface the conflict, translate both goals into measurable constraints, propose options with trade-offs (e.g., draft-only for regulated categories, auto-send for low-risk ones), and get a documented decision. D. Promise both goals will be fully met. Correct answer: C Explanation: Conflicting requirements are resolved explicitly with options and recorded decisions. Why not the others: A and B hide the conflict; D overpromises. Architectural principle: Make trade-offs explicit and decided. Ref: Prep Lesson 4
Q122 · D6 · Objective: Communicate trade-offs and set realistic expectations · Difficulty: Intermediate
Scenario: A stakeholder asks for "100% accuracy" on policy answers. Question: What is the best response? A. Agree; the model will be accurate. B. Refuse the project. C. Say accuracy cannot be discussed. D. Reframe into measurable targets (e.g., grounded-answer rate on a labeled set, escalation for low confidence), explain residual risk, and show how controls and review manage it. Correct answer: D Explanation: LLM systems are probabilistic; set measurable targets and managed residual risk rather than absolutes. Why not the others: A overpromises; B and C don't help the stakeholder decide. Architectural principle: Replace absolutes with measurable, managed risk. Ref: Prep Lesson 4
Q123 · D6 · Objective: Communicate trade-offs and set realistic expectations · Difficulty: Advanced
Scenario: Finance asks why the pilot's cost per resolved case is above projections. Analysis shows long, uncached system prompts and a top-tier model on simple intents. Question: Which message to Finance is best? A. "LLMs are expensive." B. A breakdown of cost drivers with measured evidence, plus a prioritized plan (routing simple intents to a smaller model, prompt caching, batching non-urgent work), the expected savings, and the quality guardrails verified by evals. C. "We will reduce quality to save money." D. A request for a larger budget without analysis. Correct answer: B Explanation: Credible cost communication connects drivers, actions, expected impact, and quality safeguards. Why not the others: A, C, and D lack evidence or a plan. Architectural principle: Evidence-based cost communication. Ref: Prep Lesson 4
Q124 · D6 · Objective: Build a business case and communicate value · Difficulty: Intermediate
Scenario: You must write an executive summary for a steering committee. Question: Which structure fits best? A. A detailed token-level description of the prompt. B. A model benchmark comparison. C. The problem and impact, the recommended solution at a high level, expected value with assumptions, key risks and mitigations, cost, timeline, and the decision requested. D. A list of every API feature used. Correct answer: C Explanation: Executives need the decision, value, risk, and cost, not implementation detail. Why not the others: A, B, and D are technical detail without a decision frame. Architectural principle: Lead with the decision and the value. Ref: Prep Lesson 4
Q125 · D6 · Objective: Build a business case and communicate value · Difficulty: Advanced
Scenario: A business case claims "$3M savings" based on automating 100% of tickets. Question: What is the most significant weakness? A. The currency format. B. The title. C. The use of the word "savings." D. The assumption of 100% automation and ignoring review costs, escalation, quality risk, and run costs; it should use ranges, a sensitivity analysis, and phased adoption. Correct answer: D Explanation: Unrealistic assumptions undermine credibility; honest ranges build trust. Why not the others: A–C are cosmetic. Architectural principle: Defensible assumptions and sensitivity analysis. Ref: Prep Lesson 4
Q126 · D6 · Objective: Define SLAs and manage approvals/objections · Difficulty: Advanced
Scenario: The business asks for "99.99% availability and 1-second responses" for an agent that makes multiple model and tool calls. Question: What should the architect do? A. Accept the targets. B. Decompose the request into measurable SLOs per path (e.g., p95 latency for the interactive path, with streaming; async for long tasks), identify dependencies' limits, and negotiate achievable targets with fallbacks. C. Reject all targets. D. Promise targets and fix them later. Correct answer: B Explanation: SLAs must reflect dependency behavior and be defined per workflow path. Why not the others: A and D overpromise; C avoids negotiation. Architectural principle: SLOs grounded in the architecture. Ref: Prep Lesson 4
Q127 · D6 · Objective: Define SLAs and manage approvals/objections · Difficulty: Advanced
Question (Select TWO): A security team objects that the agent could leak data. Which responses are most effective? A. Present the threat model, controls (least privilege, egress limits, output screening), and test evidence. B. Tell them the model is safe. C. Offer a limited pilot with monitoring, rollback criteria, and agreed exit conditions. D. Escalate to executives to override the objection. E. Postpone all security review until after launch. Correct answers: A, C Explanation: Evidence and a bounded, reversible rollout address the concern directly. Why not the others: B asserts; D bypasses; E defers risk. Architectural principle: Address objections with evidence and bounded risk. Ref: Prep Lesson 4
Q128 · D6 · Objective: Define SLAs and manage approvals/objections · Difficulty: Intermediate
Scenario: A change to the system prompt could alter customer-facing tone. Compliance requires approval for such changes. Question: What process fits? A. Edit production prompts directly. B. Treat prompts as versioned artifacts: change request, eval run against the regression suite, compliance sign-off, staged rollout, and rollback plan. C. Email the new prompt to the team. D. Skip approval for small edits. Correct answer: B Explanation: Prompts are production code and need change control. Why not the others: A, C, and D bypass control. Architectural principle: Version, test, approve, roll out. Ref: Prep Lesson 4
Q129 · D6 · Objective: Manage the solution lifecycle from PoC to production · Difficulty: Intermediate
Scenario: A PoC with ten curated examples impressed the sponsor, who wants to go live next week. Question: What gap should the architect highlight? A. None; the demo worked. B. The font in the demo. C. Missing production-readiness evidence: representative eval data, security review, load and failure testing, monitoring, cost model, and operational ownership. D. The model name. Correct answer: C Explanation: PoC success on curated cases does not establish production readiness. Why not the others: A ignores risk; B and D are irrelevant. Architectural principle: Exit criteria between lifecycle stages. Ref: Prep Lesson 4
Q130 · D6 · Objective: Manage the solution lifecycle from PoC to production · Difficulty: Advanced
Scenario: A pilot serves 5% of traffic. Quality is on target, but cost per case is rising with usage. Question: Before expanding to 50%, what should be done? A. Expand immediately. B. Remove monitoring. C. Lower the eval bar. D. Investigate the cost trend, apply optimizations (routing, caching, context trimming), re-validate quality, and set expansion gates on quality, cost, and incident criteria. Correct answer: D Explanation: Scale only after cost and quality gates are met. Why not the others: A, B, and C ignore the signal. Architectural principle: Gated, metric-driven rollout. Ref: Prep Lesson 4
Q131 · D6 · Objective: Manage the solution lifecycle from PoC to production · Difficulty: Advanced
Scenario: The provider announces that the model version you use will be deprecated in 90 days. Question: What is the best plan? A. Wait until it stops working. B. Switch the next day without testing. C. Plan migration now: run your eval suite on the successor, compare quality, cost, and latency, adjust prompts, canary, and keep a rollback until stable. D. Freeze the product permanently. Correct answer: C Explanation: Model changes need eval-driven migration with staged rollout. Why not the others: A and B risk outages or regressions; D is impractical. Architectural principle: Treat model version change as a release. Ref: Prep Lesson 4
Q132 · D6 · Objective: Produce documentation and handoff artifacts · Difficulty: Intermediate
Scenario: The client's team will operate the solution after launch. Question: Which handoff package is most complete? A. The source code only. B. Architecture and decision records, runbooks, eval suite and how to run it, monitoring dashboards and alerts, known limitations, security controls, cost model, and escalation contacts. C. A slide deck. D. A recorded demo. Correct answer: B Explanation: Operators need the means to run, evaluate, troubleshoot, and evolve the system. Why not the others: A, C, and D omit operations. Architectural principle: Handoff enables independent operation. Ref: Prep Lesson 4
Q133 · D6 · Objective: Produce documentation and handoff artifacts · Difficulty: Advanced
Scenario: Six months later, nobody remembers why a prompt uses a particular retrieval cutoff. Question: What documentation practice prevents this? A. Ask the original developer. B. Re-derive it each time. C. Architecture Decision Records capturing context, options, decision, trade-offs, and review date, linked to eval results. D. Comments in chat. Correct answer: C Explanation: ADRs preserve rationale and evidence. Why not the others: A–B depend on memory; D is not durable. Architectural principle: Record decisions with evidence. Ref: Prep Lesson 4
Q134 · D6 · Objective: Produce documentation and handoff artifacts · Difficulty: Intermediate
Scenario: A production-readiness review is scheduled. Question: Which item is a legitimate launch blocker? A. The logo is not final. B. Documentation uses two heading styles. C. No rollback procedure and no alerting on error rate or cost. D. A meeting is not scheduled. Correct answer: C Explanation: Without rollback and alerting, incidents cannot be contained. Why not the others: A, B, and D are minor. Architectural principle: Operability is a launch criterion. Ref: Prep Lesson 4
Q135 · D6 · Objective: Elicit and clarify business requirements through discovery · Difficulty: Advanced
Scenario: Discovery reveals the target process is deterministic, with stable rules and structured inputs. Question: What should the architect recommend? A. Use an LLM agent anyway. B. Use deterministic automation (rules or code) and reserve Claude for the unstructured parts, explaining the simpler, cheaper, more reliable option. C. Use the largest model. D. Cancel the project. Correct answer: B Explanation: Use the simplest solution that meets requirements; LLMs where ambiguity or language is the problem. Why not the others: A adds cost and risk; C is unjustified; D discards the value. Architectural principle: Simplest effective solution. Ref: Building effective agents
Q136 · D6 · Objective: Communicate trade-offs and set realistic expectations · Difficulty: Intermediate
Scenario: The sponsor wants to launch broadly after an impressive pilot, but metrics show weak performance on one customer segment. Question: What should the architect communicate? A. Only the overall average. B. That the segment is unimportant. C. Nothing, to avoid delaying. D. The segment finding with its business impact, options (targeted fix, human routing for that segment, phased launch), and a recommendation. Correct answer: D Explanation: Transparent findings with options let stakeholders decide. Why not the others: A–C hide a material risk. Architectural principle: Honest reporting with options. Ref: Prep Lesson 4
Q137 · D6 · Objective: Manage the solution lifecycle from PoC to production · Difficulty: Advanced
Question (Select TWO): Which are appropriate post-launch lifecycle practices? A. Monitor quality, cost, latency, and safety signals against agreed thresholds. B. Stop evaluating after launch. C. Feed production failures into the eval set and review them on a regular cadence. D. Never change prompts or models. E. Disable alerting to reduce noise. Correct answers: A, C Explanation: Continuous monitoring and an evolving eval set keep quality controlled. Why not the others: B, D, and E remove feedback or adaptability. Architectural principle: Operate, measure, improve. Ref: Prep Lesson 4
Q138 · D6 · Objective: Define SLAs and manage approvals/objections · Difficulty: Advanced
Scenario: A customer's contract requires that a human review all legal-risk responses within 4 hours. Question: Which design best supports this commitment? A. Hope reviewers are available. B. Route flagged items to a review queue with priority and timers, staffing coverage, escalation on aging, and dashboards tracking time-to-review against the SLA. C. Auto-approve after 4 hours. D. Remove flagged items from reporting. Correct answer: B Explanation: The SLA needs workflow, staffing, and measurement. Why not the others: A lacks controls; C violates the requirement; D hides breaches. Architectural principle: Operationalize commitments. Ref: Prep Lesson 4
Q139 · D6 · Objective: Build a business case and communicate value · Difficulty: Intermediate
Scenario: After launch, the sponsor asks, "Is it working?" Question: What should the architect present? A. Anecdotes from happy users. B. The number of API calls. C. A baseline-versus-current comparison on the agreed success metrics (resolution, quality, cost, escalation, satisfaction), with confidence and known limitations. D. A list of features. Correct answer: C Explanation: Value is shown against baselines using agreed metrics. Why not the others: A, B, and D are not outcome measures. Architectural principle: Measure outcomes against baselines. Ref: Prep Lesson 4
Q140 · D6 · Objective: Produce documentation and handoff artifacts · Difficulty: Expert
Scenario: A regulated client requires evidence that changes were reviewed, tested, and approved before release. Question: Which artifact set provides this? A. A verbal confirmation from the team. B. Release notes only. C. A chat transcript. D. A change record linking the version diff, eval results, reviewer approvals, rollout plan, and rollback outcome, stored immutably and retrievable by release. Correct answer: D Explanation: Traceable change records give auditable evidence. Why not the others: A–C are incomplete or unverifiable. Architectural principle: Auditable change management. Ref: Prep Lesson 4