Question 1
A receipt-processing pipeline always performs OCR, applies a fixed tax rule, and writes an approved record. Audit requires every stage in the same order. Which design is the simplest fit?
- Implement a deterministic workflow with explicit stage transitions.
- As an alternative, fan out specialized reviews and aggregate their findings in a coordinator.
- Assign bounded research to subagents and return concise results to the parent.
- Assess the refund threshold in ordinary application logic.
Explanation
Correct Answer
A:A fixed workflow makes ordering, validation, and retry boundaries visible in application code; an autonomous planner would add choices the requirement does not need.
Incorrect Answers
B:Parallel specialists can inspect distinct criteria independently; the parent reconciles results and applies a final decision policy; this implements independent parallel reviews rather than the demanded guaranteed processing order.
C:Separate subagent contexts retain intermediate tool output locally, so the coordinator receives only the information needed to synthesize; even if isolation of verbose research succeeds, guaranteed processing order still needs separate verification.
D:Monetary policy is deterministic and should be enforced outside natural-language generation; the model can supply context but cannot override the branch; changing deterministic business policy does not establish guaranteed processing order in this application.
Question 2
A customer-support system must decide whether to search manuals, inspect an order, or ask for clarification as new evidence arrives. What should select its next action?
- Use an agent loop that chooses tools from the current evidence.
- Apply a loop iteration budget in the orchestration code.
- Provide the reviewing agent read-only capabilities and delegate edits to an approved executor.
- As an alternative, maintain structured working state for completed steps and observations.
Explanation
Correct Answer
A:The route is not known in advance; a model-driven loop can evaluate observations and choose the next authorized step, while the harness enforces limits.
Incorrect Answers
B:Deterministic harness limits end execution independently of model intent; an instruction to stop is not a reliable execution boundary; even if bounded tool-loop steps succeeds, dynamic next-action selection still needs separate verification.
C:Splitting advisory analysis from mutation narrows the authority available during open-ended reasoning; suggestions are not themselves executable writes; for dynamic next-action selection, a control centered on advisory versus mutation boundary is not the appropriate substitute.
D:A durable task-state record lets orchestration distinguish new actions from already performed ones without depending on an ever-growing transcript; testing dynamic next-action selection would still fail after improving tracking completed actions.
Question 3
A financial firm needs a parent assistant to produce a summary without carrying raw, verbose research transcripts into its main conversation. What architecture helps?
- Allow the model to choose a sequence of diagnostic tools under bounded permissions.
- Introduce a deterministic workflow with explicit stage transitions.
- Delegate bounded research to subagents and return concise results to the parent.
- Provide exploratory subtasks separate context and feed bounded summaries to the coordinator.
Explanation
Correct Answer
C:Separate subagent contexts retain intermediate tool output locally, so the coordinator receives only the information needed to synthesize.
Incorrect Answers
A:Because observations determine subsequent steps, a controlled agent can adapt its plan while the harness constrains its scope and cost; it improves exploratory diagnostic sequence only, leaving isolation of verbose research unproven.
B:A fixed workflow makes ordering, validation, and retry boundaries visible in application code; an autonomous planner would add choices the requirement does not need; for isolation of verbose research, this instead manages guaranteed processing order; the requested decision remains unresolved.
D:Isolating exploratory traces protects the main context from unrelated evidence while retaining task-relevant conclusions; the effect concerns preserving coordinator focus; the success condition is isolation of verbose research.
Question 4
An automation has five known verification steps, but the team proposes a free-form agent to improve flexibility. What concern should drive the design review?
- Avoid agent autonomy when a predetermined sequence already meets the requirements.
- Require a loop iteration budget in the orchestration code.
- Designate a supervisor responsible for delegation and termination.
- Cross-check evidence and uncertainty before producing its final synthesis.
Explanation
Correct Answer
A:An agent introduces nondeterministic planning and extra testing surface without improving a process whose branches can be specified directly.
Incorrect Answers
B:Deterministic harness limits end execution independently of model intent; an instruction to stop is not a reliable execution boundary; testing unnecessary autonomous decisions would still fail after improving bounded tool-loop steps.
C:A supervisor can track completion criteria and prevent uncontrolled peer handoffs; the workflow has one accountable termination decision; for this request, unnecessary autonomous decisions is the missing piece, not controlled delegation ownership.
D:Parallel delegation accelerates collection, but agreement does not arise automatically; the coordinator must evaluate conflicts and decide whether to request more evidence; it changes evidence reconciliation; it does not prove unnecessary autonomous decisions in this case.
Question 5
A travel assistant can search, compare, and book, but payment must occur only after explicit user acceptance. Where should authority for booking reside?
- Place a deterministic approval gate between planning and the booking tool.
- Require a concrete benefit from independent context or genuinely parallel subproblems.
- Grant the reviewing agent read-only capabilities and delegate edits to an approved executor.
- Build a deterministic workflow with explicit stage transitions.
- As an alternative, avoid agent autonomy when a predetermined sequence already meets the requirements.
Explanation
Correct Answer
A:The agent may draft an itinerary, but a separate permission check must decide whether the consequential booking invocation is authorized.
Incorrect Answers
B:Extra agents add coordination, latency, and failure modes; they earn their cost only when workload structure provides useful separation; testing user-approved irreversible action would still fail after improving multi-criterion integration.
C:Splitting advisory analysis from mutation narrows the authority available during open-ended reasoning; suggestions are not themselves executable writes; it improves advisory versus mutation boundary only, leaving user-approved irreversible action unproven.
D:A fixed workflow makes ordering, validation, and retry boundaries visible in application code; an autonomous planner would add choices the requirement does not need; changing guaranteed processing order does not establish user-approved irreversible action in this application.
E:An agent introduces nondeterministic planning and extra testing surface without improving a process whose branches can be specified directly; where user-approved irreversible action matters, a control for unnecessary autonomous decisions leaves that test unmet.
Question 6
A software-maintenance manager wants independent security and performance analysis on the same pull request before issuing one decision. What pattern matches this work?
- Hand off bounded research to subagents and return concise results to the parent.
- Grant exploratory subtasks separate context and feed bounded summaries to the coordinator.
- Fan out specialized reviews and aggregate their findings in a coordinator.
- Permit the model to choose a sequence of diagnostic tools under bounded permissions.
Explanation
Correct Answer
C:Parallel specialists can inspect distinct criteria independently; the parent reconciles results and applies a final decision policy.
Incorrect Answers
A:Separate subagent contexts retain intermediate tool output locally, so the coordinator receives only the information needed to synthesize; it changes isolation of verbose research; it does not prove independent parallel reviews in this case.
B:Isolating exploratory traces protects the main context from unrelated evidence while retaining task-relevant conclusions; where independent parallel reviews matters, a control for preserving coordinator focus leaves that test unmet.
D:Because observations determine subsequent steps, a controlled agent can adapt its plan while the harness constrains its scope and cost; this targets exploratory diagnostic sequence, while the system must demonstrate independent parallel reviews.
Question 7
A supervisor delegates evidence gathering to three incident workers. One worker times out, while the other two return usable findings. What should the supervisor do before answering?
- Treat a worker timeout as confirmation that the missing hypothesis was false.
- Automatically rerun every successful worker and discard all previous findings.
- Invent the missing worker report to make the coordinator summary structurally complete.
- Mark the timed-out branch incomplete, reconcile the available findings, and disclose the missing evidence.
Explanation
Correct Answer
D:A failed delegated branch is not a successful observation. The supervisor can continue with the verified outputs when policy permits, but must preserve the failed branch status and qualify any conclusion that depended on it.
Incorrect Answers
A:Absence of a tool or specialist result provides no negative evidence about the hypothesis. Claiming a finding from a timeout would convert an execution failure into an unsupported technical conclusion.
B:Replaying completed investigations wastes calls and may introduce inconsistent observations. Recovery should target the failed branch, with bounded retries and a clear partial-result policy.
C:A complete-looking format cannot replace evidence that was never obtained. The coordinator should represent missing work explicitly rather than fabricate verification outcomes.
Question 8
A review agent has broad editing authority although the task only requires suggestions. What architecture change best reduces accidental writes?
- Assign a supervisor responsible for delegation and termination.
- Assign exploratory subtasks separate context and feed bounded summaries to the coordinator.
- Give the reviewing agent read-only capabilities and delegate edits to an approved executor.
- Put a deterministic approval gate between planning and the booking tool.
Explanation
Correct Answer
C:Splitting advisory analysis from mutation narrows the authority available during open-ended reasoning; suggestions are not themselves executable writes.
Incorrect Answers
A:A supervisor can track completion criteria and prevent uncontrolled peer handoffs; the workflow has one accountable termination decision; even if controlled delegation ownership succeeds, advisory versus mutation boundary still needs separate verification.
B:Isolating exploratory traces protects the main context from unrelated evidence while retaining task-relevant conclusions; the advisory versus mutation boundary criterion differs from preserving coordinator focus; the gap persists.
D:The agent may draft an itinerary, but a separate permission check must decide whether the consequential booking invocation is authorized; where advisory versus mutation boundary matters, a control for user-approved irreversible action leaves that test unmet.
Question 9
A manager agent delegates web research to three workers, but their outputs contradict one another. What must the manager do?
- Position a deterministic approval gate between planning and the booking tool.
- Reconcile evidence and uncertainty before producing its final synthesis.
- For this workflow, fan out specialized reviews and aggregate their findings in a coordinator.
- Mandate a concrete benefit from independent context or genuinely parallel subproblems.
Explanation
Correct Answer
B:Parallel delegation accelerates collection, but agreement does not arise automatically; the coordinator must evaluate conflicts and decide whether to request more evidence.
Incorrect Answers
A:The agent may draft an itinerary, but a separate permission check must decide whether the consequential booking invocation is authorized; testing evidence reconciliation would still fail after improving user-approved irreversible action.
C:Parallel specialists can inspect distinct criteria independently; the parent reconciles results and applies a final decision policy; because independent parallel reviews is its concern, evidence reconciliation remains outside its effect here.
D:Extra agents add coordination, latency, and failure modes; they earn their cost only when workload structure provides useful separation; for evidence reconciliation, a control centered on multi-criterion integration is not the appropriate substitute.
Question 10
A customer workflow requires a strict maximum of six model-tool cycles per request. Where should the limit be implemented?
- Name a supervisor responsible for delegation and termination.
- For this workflow, avoid agent autonomy when a predetermined sequence already meets the requirements.
- Create an explicit draft-critique workflow with a defined exit condition.
- Enforce a loop iteration budget in the orchestration code.
Explanation
Correct Answer
D:Deterministic harness limits end execution independently of model intent; an instruction to stop is not a reliable execution boundary.
Incorrect Answers
A:A supervisor can track completion criteria and prevent uncontrolled peer handoffs; the workflow has one accountable termination decision; testing bounded tool-loop steps would still fail after improving controlled delegation ownership.
B:An agent introduces nondeterministic planning and extra testing surface without improving a process whose branches can be specified directly; for bounded tool-loop steps, a control centered on unnecessary autonomous decisions is not the appropriate substitute.
C:A known two-stage procedure is easier to test as a fixed workflow than as an unrestricted multi-agent conversation; a remedy for bounded critique stage is not a substitute for bounded tool-loop steps.