Question 90
Two independent checks take 4 and 6 seconds, followed by a 2-second synthesis. Ignoring overhead, what is the shortest elapsed time if both checks must finish?
- 8 seconds
- 12 seconds
- 6 seconds
- 2 seconds
Explanation
Correct Answer
A:Parallel checks complete when the slower six-second check finishes. The dependent two-second synthesis then makes the total eight seconds.
Incorrect Answers
B:This adds both independent check durations as though they were sequential. Concurrency allows their execution times to overlap.
C:This omits the synthesis stage that cannot start until both checks finish. The longest check is not the whole request path.
D:Synthesis depends on both check results. Its duration alone cannot represent completion of the required end-to-end workflow.
Question 91
A planner explores every combination of twelve optional analysis steps. Many combinations violate known compatibility rules, but they are rejected only after full evaluation. Which change most directly reduces wasted search?
- Evaluate all combinations with identical priority
- Choose a branch only by its position in the list
- Evaluate every branch to completion before rejecting infeasible ones
- Prune infeasible branches using explicit task constraints
Explanation
Correct Answer
D:Requirements can eliminate combinations that cannot contribute to a valid result. Constraint-guided planning reduces wasted exploration without discarding useful paths arbitrarily.
Incorrect Answers
A:Equal priority preserves the explosion and ignores information about feasibility. A planner should use constraints to focus its effort.
B:Position does not establish suitability. Arbitrary selection reduces work by abandoning the requirement to choose a feasible plan.
C:Late pruning spends work on paths already excluded by explicit constraints. Early feasibility checks prevent unnecessary expansion of the planning search.
Question 92
A support assistant can search manuals, draft replies, and close customer accounts. Which autonomy policy best matches the differing consequences?
- Allow bounded search while separately restricting account closure
- Remove all tools because one is destructive
- Give all tools the same broad automatic-execution policy
- Require identical approval for every retrieved paragraph
Explanation
Correct Answer
A:Autonomy can vary by capability rather than apply uniformly. Read-only investigation and destructive changes require different authority and review conditions.
Incorrect Answers
B:The architecture can preserve useful low-risk work while constraining the dangerous action. Removing every capability is unnecessarily broad for the stated needs.
C:Uniform permissions ignore the much greater consequence of account closure. The low risk of one tool does not justify another tool's authority.
D:Uniform review of trivial reads creates friction without targeting the highest-risk operation. Controls should reflect the consequences of each capability.
Question 93
An assistant's own generated answers are automatically added to its knowledge base as verified facts. Accuracy drifts downward. What architecture change is needed?
- Validate feedback before admitting it as authoritative knowledge
- Use the same model to approve every answer automatically
- Remove timestamps from generated entries
- Increase retrieval priority for frequently repeated answers
Explanation
Correct Answer
A:Generated content is not independent evidence of truth. Unchecked ingestion can reinforce errors through repeated retrieval and make their origin harder to detect.
Incorrect Answers
B:An uncalibrated self-check may repeat the original mistake. Admission needs an appropriate verification process tied to reliable evidence and risk.
C:Removing provenance makes contamination harder to investigate. The system needs stronger admission and traceability controls, not less context about origins.
D:Repetition can be caused by the feedback loop itself. Promoting repeated material may amplify the same unsupported claim further.
Question 94
An agent promises next-day processing, but the service operates only on business days in the customer's timezone. What should determine the promised date?
- A calendar-aware scheduling component with the relevant timezone
- The server's UTC date without conversion
- The model's general interpretation of tomorrow
- The most common date in past responses
Explanation
Correct Answer
A:The commitment depends on explicit operational rules and local dates. A validated scheduling function can apply weekends, holidays, and cutoff times consistently.
Incorrect Answers
B:UTC is useful for storage but may differ from the customer's business date. Scheduling must apply the timezone specified by the service contract.
C:General language does not encode the organization's calendar or cutoff rules. A plausible date can still be an invalid service commitment.
D:Historical frequency cannot establish which date applies to this request. Holidays and cutoffs require current rule-based evaluation.
Question 95
Temporary analysis environments remain active after agent jobs fail, causing ongoing cost. Which lifecycle design directly addresses this?
- Let only successful jobs delete their environments
- Assign cleanup responsibility for success, failure, and cancellation
- Ask users to find and delete unknown environments
- Increase the idle timeout to avoid interrupting work that may still be running
Explanation
Correct Answer
B:Resource ownership must include termination paths, not just normal completion. A cleanup mechanism should reclaim temporary environments when work ends abnormally too.
Incorrect Answers
A:This leaves the exact failed-job leak unresolved. Cleanup must cover exceptional paths as well as successful execution.
C:Users may not know which resources were created or whether deletion is safe. The system that allocates them needs a reliable lifecycle contract.
D:A longer timeout can increase the cost of abandoned environments. Cleanup needs to recognize terminal job states and release resources appropriately.
Question 96
A coordinator delegates research through several layers and receives a conclusion without its original sources. What must be preserved through each handoff?
- Evidence lineage linking conclusions to originating artifacts
- Only a shared confidence score for the whole job
- Only the coordinator's final consolidated conclusion
- Only the identity of the last agent in the chain
Explanation
Correct Answer
A:Delegation can compress findings while retaining traceable evidence. Without lineage, later stages cannot distinguish supported conclusions from unsupported summaries.
Incorrect Answers
B:A global score loses claim-level distinctions. It cannot identify which source supports a specific conclusion or where uncertainty entered.
C:A conclusion without its evidence links cannot explain which worker and artifact supported each claim. Delegated provenance must survive consolidation.
D:The last agent may merely repeat a claim. Its identity does not establish where the underlying evidence originated or how it was transformed.
Question 97
Runtime controls enforce a maximum tool-call count and block writes in read-only workflows. Which release check most directly detects regressions in these constraints on the tested paths?
- Automated constraint checks on representative execution traces
- Compare only the aggregate number of tool calls across the entire test suite
- Check tool schemas without executing any workflow paths
- A diagram reviewed only during initial planning
Explanation
Correct Answer
A:Trace assertions can detect excess calls and prohibited write attempts in exercised paths. They complement the runtime controls; passing a finite test suite does not prove that every possible execution is safe.
Incorrect Answers
B:An aggregate can conceal one workflow exceeding its limit while another makes fewer calls. It also cannot establish whether a read-only path attempted a write.
C:Schemas describe call shapes rather than the sequence of calls a workflow makes. They do not reveal excessive calls or prohibited write attempts during execution.
D:An initial diagram can explain intent but cannot detect later behavior drift. The constraints concern the running implementation after changes.
Question 98
A worker adds a new required field to its output, but an older coordinator cannot interpret it. Which rollout strategy best reduces integration breakage?
- Remove schema validation until every component updates
- Version the contract and coordinate compatible producer-consumer changes
- Deploy the worker first and let the coordinator infer meaning
- Replace the producer contract first and update consumers when errors appear
Explanation
Correct Answer
B:A required-field change affects both sides of the handoff. Versioning and compatibility testing make the transition explicit rather than assuming immediate coordinated adoption.
Incorrect Answers
A:Disabling validation hides incompatibility and can allow corrupt state downstream. Compatibility needs deliberate handling, not removal of the signal.
C:Inference cannot reliably replace a machine contract. The old consumer may reject the result or silently interpret it incorrectly.
D:That deliberately creates an incompatibility window. Coordinated versioning or a compatible transition is needed before consumers depend on the new field.
Question 99
A bulk update contains 500 proposed changes. Operators need to inspect the affected records before any write occurs. Which tool design best supports this?
- A single tool that writes and then lists changed records
- A write tool that stops after the first error
- A prompt asking the model to describe likely records
- A dry-run preparation phase followed by explicit commit
Explanation
Correct Answer
D:Preparation can calculate and expose the proposed impact without applying it. A separate commit boundary lets the approved set govern execution.
Incorrect Answers
A:Post-write reporting reveals impact after the consequence. It cannot satisfy inspection before the first record is changed.
B:Stopping on error limits some damage but does not provide pre-execution inspection. The requirement concerns review of the full proposed update.
C:A likely set is not a verified impact preview. The preparation phase should resolve actual targets using the same relevant selection rules.
Question 100
A business workflow may pause for a manager response lasting several days. Workers are scarce, and all information needed to resume can be persisted. Which design best handles the wait?
- Persist a resumable state and reactivate execution when the response arrives
- Poll the manager's inbox once per second from every paused worker
- Keep a worker thread blocked for the entire approval period
- Restart the entire workflow from its first step after every response
Explanation
Correct Answer
A:The workflow does not need an occupied worker while no computation can proceed. A durable wait state preserves continuity and permits event-driven resumption.
Incorrect Answers
B:Frequent per-worker polling adds unnecessary load and retains coupling to active workers. A bounded event-driven resumption contract better fits the long wait.
C:A blocked thread consumes execution capacity without doing useful work. The multi-day wait can be represented durably outside the active worker.
D:A full restart discards persisted progress and may repeat completed actions. The scenario allows continuation from a stored state, which avoids that unnecessary work.