Question 40
A research orchestrator keeps creating new subagents for minor subtasks until its token budget is exhausted. Which approach should be recommended?
- Validate worker contracts at handoff and quarantine malformed worker output.
- Send only task-specific context and concise evidence references to each worker.
- Cap delegation depth, child count and per-job resource budgets.
- Use isolated agent workspaces and explicit structured handoff artifacts.
Explanation
Correct Answer
C:Structured limits protect cost and latency without assuming the model will spontaneously stop exploring. Assign a coordinator ownership of controlled multi-agent resource use and retain evidence and permission boundaries at each specialist handoff.
Incorrect Answers
A:The option might improve fault isolation at agent interfaces; that remains insufficient for controlled multi-agent resource use. The case is resolved by controlled multi-agent resource use, which depends on a design that will cap delegation depth, child count and per-job resource budgets.
B:The proposed action prioritizes minimal agent handoff context. That can coexist with the right solution, but it cannot replace controlled multi-agent resource use: the application must still cap delegation depth, child count and per-job resource budgets.
D:This addresses isolated state with accountable handoffs; however, the problem asks for controlled multi-agent resource use. Without the safeguard to cap delegation depth, child count and per-job resource budgets, the original failure mode is still possible.
Question 41
Agents can finish independently, but the final answer sometimes appears before the policy specialist returns. What is the best production design response?
- Use an orchestrator to delegate scoped investigations to specialist agents and synthesize a reviewed conclusion.
- Use isolated agent workspaces and explicit structured handoff artifacts.
- Record side-effect completion before retries and route uncertain outcomes to reconciliation.
- Require explicit completion or timeout reconciliation before final synthesis.
Explanation
Correct Answer
D:The coordinator must account for missing branches and show partial confidence rather than silently omitting mandatory work.
Incorrect Answers
A:That is primarily a technique for bounded specialist delegation. It lacks the explicit measure for complete and timed aggregation that is necessary here, namely a system that will require explicit completion or timeout reconciliation before final synthesis.
B:Improving isolated state with accountable handoffs does not imply complete and timed aggregation. The selected design must require explicit completion or timeout reconciliation before final synthesis, whereas this choice omits that direct protection or verification step.
C:This recommendation targets transaction safety across agent retries at a different point in the workflow. It leaves complete and timed aggregation outside the design contract and does not require the team to require explicit completion or timeout reconciliation before final synthesis.
Question 42
A team uses a costly group of specialized agents to answer trivial FAQs. In its real investigation workflow, one worker forwards text from an untrusted web page as if it were a coordinator instruction, triggering inappropriate delegated tool calls. Which TWO corrections are needed? Choose TWO.
- Cap delegation depth, child count and per-job resource budgets.
- Use an orchestrator to delegate scoped investigations to specialist agents and synthesize a reviewed conclusion.
- Log task assignments, tool calls, evidence references and synthesis decisions using shared trace identifiers.
- Treat worker findings as untrusted data and independently authorize delegated tool calls at the coordinator boundary.
- Use one grounded assistant until evaluation shows a clear benefit from decomposition.
Explanation
Correct Answers
D:A specialist report is evidence, not a new source of executable authority. The coordinator must distinguish source text from instructions and validate the tool, parameters, and permitted task before executing any delegated action.
E:A single grounded assistant is an adequate baseline for straightforward FAQs. Specialist coordination adds cost, state boundaries and opportunities for tool misuse without evidence of better answers; retaining many agents is justified only if representative evaluation shows a material gain.
Incorrect Answers
A:The control would be justified if controlled multi-agent resource use were the limiting factor. In this scenario the decisive target is avoiding unnecessary multi-agent complexity and trust-boundary enforcement at interagent handoffs; success requires an architecture that will treat worker findings as untrusted data and independently authorize delegated tool calls at the coordinator boundary.
B:The option belongs to the broader work on bounded specialist delegation, but it cannot by itself deliver avoiding unnecessary multi-agent complexity and trust-boundary enforcement at interagent handoffs. A decision on this case needs the team to use one grounded assistant until evaluation shows a clear benefit from decomposition and verify that outcome.
C:This shifts attention toward auditability of distributed reasoning when the relevant decision is about avoiding unnecessary multi-agent complexity and trust-boundary enforcement at interagent handoffs. Unless the architecture can treat worker findings as untrusted data and independently authorize delegated tool calls at the coordinator boundary, the failure described in the stem remains unresolved.
Question 43
A supervisor agent retries a failed worker three times and then repeats a costly external write. What design decision would best fit these facts?
- Record side-effect completion before retries and route uncertain outcomes to reconciliation.
- Require explicit completion or timeout reconciliation before final synthesis.
- Bind tools and credentials to each worker's approved task scope.
- Provision a scoped credential solely to that agent and pass a minimized result to the coordinator.
Explanation
Correct Answer
A:At-least-once agent invocation is unsafe for non-idempotent operations unless commit state is tracked. Assign a coordinator ownership of transaction safety across agent retries and retain evidence and permission boundaries at each specialist handoff.
Incorrect Answers
B:That choice would be responsive to complete and timed aggregation, but the reported defect is different. It leaves transaction safety across agent retries unproved because there is no commitment to record side-effect completion before retries and route uncertain outcomes to reconciliation.
C:There is a legitimate use case for per-agent permission boundaries, but it does not supply the test or enforcement needed for transaction safety across agent retries. The design needs to record side-effect completion before retries and route uncertain outcomes to reconciliation.
D:This addresses data minimization across specialist agents; however, the problem asks for transaction safety across agent retries. Without the safeguard to record side-effect completion before retries and route uncertain outcomes to reconciliation, the original failure mode is still possible.
Question 44
An organization needs to audit why the summarizer accepted one agent's finding and rejected another's. Which intervention most directly addresses the failure?
- Log task assignments, tool calls, evidence references and synthesis decisions using shared trace identifiers.
- Use isolated agent workspaces and explicit structured handoff artifacts.
- Use an orchestrator to delegate scoped investigations to specialist agents and synthesize a reviewed conclusion.
- Send only task-specific context and concise evidence references to each worker.
Explanation
Correct Answer
A:A joined execution trace exposes cross-agent dependencies and allows disputed outputs to be reconstructed. Assign a coordinator ownership of auditability of distributed reasoning and retain evidence and permission boundaries at each specialist handoff.
Incorrect Answers
B:This would make sense as an additional project activity for isolated state with accountable handoffs. The priority required here is auditability of distributed reasoning, so the workflow still needs a control that will log task assignments, tool calls, evidence references and synthesis decisions using shared trace identifiers.
C:By itself, this improves bounded specialist delegation, but the case is not asking the architect to optimize that dimension. The requirement is auditability of distributed reasoning, supported by an approach to log task assignments, tool calls, evidence references and synthesis decisions using shared trace identifiers.
D:Its main effect concerns minimal agent handoff context. The case explicitly hinges on auditability of distributed reasoning; to resolve it reliably, the implementation has to log task assignments, tool calls, evidence references and synthesis decisions using shared trace identifiers, which this choice does not establish.
Question 45
A customer-impact agent needs an external database that security forbids other specialists to query. What should the solution architect prioritize?
- Use an orchestrator to delegate scoped investigations to specialist agents and synthesize a reviewed conclusion.
- Use isolated agent workspaces and explicit structured handoff artifacts.
- Record side-effect completion before retries and route uncertain outcomes to reconciliation.
- Provision a scoped credential solely to that agent and pass a minimized result to the coordinator.
Explanation
Correct Answer
D:A specialist may have a necessary sensitive integration while other agents only receive authorized derived facts.
Incorrect Answers
A:For a workflow whose primary weakness was bounded specialist delegation, that proposal could be reasonable. Here the unmet obligation is data minimization across specialist agents, and that requires the system to provision a scoped credential solely to that agent and pass a minimized result to the coordinator.
B:Although isolated state with accountable handoffs can be valuable, it is not the acceptance condition in the question. The missing provision is data minimization across specialist agents, addressed by the control to provision a scoped credential solely to that agent and pass a minimized result to the coordinator.
C:It is aimed at a different outcome, transaction safety across agent retries. No evidence is given that it provides data minimization across specialist agents; the needed mechanism is to provision a scoped credential solely to that agent and pass a minimized result to the coordinator at the appropriate boundary.
Question 46
An orchestrator receives partially invalid structured responses from specialists and crashes during merger. Which design choice best matches the stated constraint?
- Use isolated agent workspaces and explicit structured handoff artifacts.
- Validate worker contracts at handoff and quarantine malformed worker output.
- Bind tools and credentials to each worker's approved task scope.
- Cap delegation depth, child count and per-job resource budgets.
Explanation
Correct Answer
B:Typed delegation results keep one errant worker from corrupting the entire aggregate response. Assign a coordinator ownership of fault isolation at agent interfaces and retain evidence and permission boundaries at each specialist handoff.
Incorrect Answers
A:Consider its actual operating scope: isolated state with accountable handoffs. The requested assurance is fault isolation at agent interfaces, which calls for an implementation that can validate worker contracts at handoff and quarantine malformed worker output, not merely for a related improvement.
C:The proposed measure is aimed at per-agent permission boundaries, not fault isolation at agent interfaces. The architecture must instead validate worker contracts at handoff and quarantine malformed worker output; a separate change is still needed for fault isolation at agent interfaces.
D:The relevant trade-off is not primarily controlled multi-agent resource use. A successful fix needs fault isolation at agent interfaces; otherwise fault isolation at agent interfaces remains unverified even after this change.
Question 47
A contract-review job contains extraction, clause matching and risk assessment; assessment assumes verified extraction. Which architecture adjustment is most appropriate?
- Fan out to both reviewers, then gate publication on their combined approved outcomes.
- Partition the listings and run checks in parallel with rate-limited workers.
- Persist intermediate artifacts and invalidate only affected stages.
- Use sequential steps with validation checkpoints between dependent stages.
- Break the task into typed steps with externally verified computations.
- Summarize sources in parallel and perform a final cross-source reconciliation stage.
Explanation
Correct Answer
D:Later stages rely on upstream facts, so ordering protects correctness more than parallel execution would. Keep the dependency or branch affecting dependency-aware task chaining observable, so a failed subtask cannot be mistaken for a completed workflow.
Incorrect Answers
A:Its main effect concerns fan-out review with synchronized approval. The case explicitly hinges on dependency-aware task chaining; to resolve it reliably, the implementation has to use sequential steps with validation checkpoints between dependent stages, which this choice does not establish.
B:The option might improve rate-limited parallel decomposition; that remains insufficient for dependency-aware task chaining. The case is resolved by dependency-aware task chaining, which depends on a design that will use sequential steps with validation checkpoints between dependent stages.
C:The proposed action prioritizes selective workflow recomputation. That can coexist with the right solution, but it cannot replace dependency-aware task chaining: the application must still use sequential steps with validation checkpoints between dependent stages.
E:This addresses inspectable modular calculation; however, the problem asks for dependency-aware task chaining. Without the safeguard to use sequential steps with validation checkpoints between dependent stages, the original failure mode is still possible.
F:That choice would be responsive to parallel summarization with final consistency checking, but the reported defect is different. It leaves dependency-aware task chaining unproved because there is no commitment to use sequential steps with validation checkpoints between dependent stages.
Question 48
A large catalog contains thousands of independent listings needing the same safety check. Which next step is most defensible?
- Use sequential steps with validation checkpoints between dependent stages.
- Partition the listings and run checks in parallel with rate-limited workers.
- Fan out to both reviewers, then gate publication on their combined approved outcomes.
- Split the work by accountable outputs, then integrate with explicit contracts.
Explanation
Correct Answer
B:Independent units permit concurrency while quotas and aggregation protect service reliability. Keep the dependency or branch affecting rate-limited parallel decomposition observable, so a failed subtask cannot be mistaken for a completed workflow.
Incorrect Answers
A:It emphasizes dependency-aware task chaining without closing the stated gap in rate-limited parallel decomposition. By contrast, the remedy to partition the listings and run checks in parallel with rate-limited workers acts on the condition that makes this scenario fail.
C:Improving fan-out review with synchronized approval does not imply rate-limited parallel decomposition. The selected design must partition the listings and run checks in parallel with rate-limited workers, whereas this choice omits that direct protection or verification step.
D:This recommendation targets work packages with verifiable interfaces at a different point in the workflow. It leaves rate-limited parallel decomposition outside the design contract and does not require the team to partition the listings and run checks in parallel with rate-limited workers.
Question 49
A support platform needs distinct workflows for refunds, technical troubleshooting and shipment inquiries. Which approach should be recommended?
- Remove the unnecessary debate step and retest quality under the target load.
- Persist intermediate artifacts and invalidate only affected stages.
- Classify the request and route to a domain-specific workflow with constrained tools.
- Mark mandatory and optional branches, then degrade with evidence of omitted sources.
- Use a bounded generate-critique-revise pattern tied to acceptance criteria.
Explanation
Correct Answer
C:Routing prevents a universal prompt from carrying unnecessary capabilities and conflicting instructions. Keep the dependency or branch affecting intent-driven branch selection observable, so a failed subtask cannot be mistaken for a completed workflow.
Incorrect Answers
A:The relevant trade-off is not primarily optimization of workflow complexity. A successful fix needs intent-driven branch selection; otherwise intent-driven branch selection remains unverified even after this change.
B:The control would be justified if selective workflow recomputation were the limiting factor. In this scenario the decisive target is intent-driven branch selection; success requires an architecture that will classify the request and route to a domain-specific workflow with constrained tools.
D:This shifts attention toward explicit partial-result semantics when the relevant decision is about intent-driven branch selection. Unless the architecture can classify the request and route to a domain-specific workflow with constrained tools, the failure described in the stem remains unresolved.
E:For a workflow whose primary weakness was controlled evaluation-driven revision, that proposal could be reasonable. Here the unmet obligation is intent-driven branch selection, and that requires the system to classify the request and route to a domain-specific workflow with constrained tools.
Question 50
A report draft can be refined against a measurable quality rubric but should not loop indefinitely. Which change most directly satisfies this requirement?
- Classify the request and route to a domain-specific workflow with constrained tools.
- Fan out to both reviewers, then gate publication on their combined approved outcomes.
- Remove the unnecessary debate step and retest quality under the target load.
- Use a bounded generate-critique-revise pattern tied to acceptance criteria.
Explanation
Correct Answer
D:Evaluator feedback supplies specific corrections while a retry ceiling prevents endless polishing. Keep the dependency or branch affecting controlled evaluation-driven revision observable, so a failed subtask cannot be mistaken for a completed workflow.
Incorrect Answers
A:The design concern is real but different: intent-driven branch selection. It does not impose the specific requirement for controlled evaluation-driven revision; a suitable solution must use a bounded generate-critique-revise pattern tied to acceptance criteria and demonstrate the result.
B:This addresses fan-out review with synchronized approval; however, the problem asks for controlled evaluation-driven revision. Without the safeguard to use a bounded generate-critique-revise pattern tied to acceptance criteria, the original failure mode is still possible.
C:That choice would be responsive to optimization of workflow complexity, but the reported defect is different. It leaves controlled evaluation-driven revision unproved because there is no commitment to use a bounded generate-critique-revise pattern tied to acceptance criteria.