Question 140
An interactive support assistant makes three serial retrieval calls although all evidence queries are independent. What design decision would best fit these facts?
- Cache authorized stable retrieval results with version and TTL validation.
- Adopt separate fast and deep-service paths with explicit deadlines and fallback behavior.
- Run independent reads concurrently and merge results with bounded timeouts.
- Use exponential backoff with jitter and a bounded retry budget for transient failures.
- Run accuracy-versus-latency evaluations across representative hard and easy queries.
Explanation
Correct Answer
C:Parallel reads reduce critical-path latency when results have no dependency on one another. Measure concurrent evidence retrieval under representative concurrency and source freshness constraints; low latency is not success when evidence for concurrent evidence retrieval is missing.
Incorrect Answers
A:This is a valid architectural concern when optimizing safe caching of stable evidence. Here it leaves concurrent evidence retrieval unaddressed, whereas the intended change must run independent reads concurrently and merge results with bounded timeouts before the case can be considered resolved.
B:This would make sense as an additional project activity for quality-aware deadline management. The priority required here is concurrent evidence retrieval, so the workflow still needs a control that will run independent reads concurrently and merge results with bounded timeouts.
D:Its main effect concerns throttling-aware retry policy. The case explicitly hinges on concurrent evidence retrieval; to resolve it reliably, the implementation has to run independent reads concurrently and merge results with bounded timeouts, which this choice does not establish.
E:The option might improve validated speed and recall tradeoff; that remains insufficient for concurrent evidence retrieval. The case is resolved by concurrent evidence retrieval, which depends on a design that will run independent reads concurrently and merge results with bounded timeouts.
Question 141
A financial agent fetches real-time balances on every prompt, including unrelated greetings. What should the solution architect prioritize?
- Run independent reads concurrently and merge results with bounded timeouts.
- Trigger high-cost tool calls only when the task needs current account data.
- Apply reranking selectively to ambiguous or low-confidence retrievals.
- Separate rarely changing descriptive text from authoritative real-time price retrieval.
- Place integration gateways near supported workloads while respecting the data residency boundary.
- Use lazy tool discovery and call only capabilities required for the resolved intent.
Explanation
Correct Answer
B:Selective invocation removes irrelevant network round trips while protecting correctness for balance questions.
Incorrect Answers
A:For a workflow whose primary weakness was concurrent evidence retrieval, that proposal could be reasonable. Here the unmet obligation is relevance-driven tool execution, and that requires the system to trigger high-cost tool calls only when the task needs current account data.
C:It is aimed at a different outcome, adaptive retrieval quality cost. No evidence is given that it provides relevance-driven tool execution; the needed mechanism is to trigger high-cost tool calls only when the task needs current account data at the appropriate boundary.
D:It emphasizes field-specific freshness guarantees without closing the stated gap in relevance-driven tool execution. By contrast, the remedy to trigger high-cost tool calls only when the task needs current account data acts on the condition that makes this scenario fail.
E:That is primarily a technique for locality-aware integration topology. It lacks the explicit measure for relevance-driven tool execution that is necessary here, namely a system that will trigger high-cost tool calls only when the task needs current account data.
F:Improving on-demand capability loading does not imply relevance-driven tool execution. The selected design must trigger high-cost tool calls only when the task needs current account data, whereas this choice omits that direct protection or verification step.
Question 142
A product search system answers quickly using a stale cached index, but price changes must appear immediately. Which design choice best matches the stated constraint?
- Separate rarely changing descriptive text from authoritative real-time price retrieval.
- Cache authorized stable retrieval results with version and TTL validation.
- Adopt separate fast and deep-service paths with explicit deadlines and fallback behavior.
- Apply reranking selectively to ambiguous or low-confidence retrievals.
Explanation
Correct Answer
A:Freshness requirements determine which fields may be cached and which require live source queries. Measure field-specific freshness guarantees under representative concurrency and source freshness constraints; low latency is not success when evidence for field-specific freshness guarantees is missing.
Incorrect Answers
B:The alternative could reduce risk associated with safe caching of stable evidence, yet the described behavior requires field-specific freshness guarantees. That outcome depends on the capability to separate rarely changing descriptive text from authoritative real-time price retrieval.
C:The proposed measure is aimed at quality-aware deadline management, not field-specific freshness guarantees. The architecture must instead separate rarely changing descriptive text from authoritative real-time price retrieval; a separate change is still needed for field-specific freshness guarantees.
D:The relevant trade-off is not primarily adaptive retrieval quality cost. A successful fix needs field-specific freshness guarantees; otherwise field-specific freshness guarantees remains unverified even after this change.
Question 143
A customer chat repeatedly queries the same stable product policy within a few minutes. Which architecture adjustment is most appropriate?
- Separate rarely changing descriptive text from authoritative real-time price retrieval.
- Cache authorized stable retrieval results with version and TTL validation.
- Run accuracy-versus-latency evaluations across representative hard and easy queries.
- Run independent reads concurrently and merge results with bounded timeouts.
Explanation
Correct Answer
B:Short-lived retrieval caching can reduce latency without treating volatile customer state as immutable. Measure safe caching of stable evidence under representative concurrency and source freshness constraints; low latency is not success when evidence for safe caching of stable evidence is missing.
Incorrect Answers
A:Its main effect concerns field-specific freshness guarantees. The case explicitly hinges on safe caching of stable evidence; to resolve it reliably, the implementation has to cache authorized stable retrieval results with version and TTL validation, which this choice does not establish.
C:The proposed action prioritizes validated speed and recall tradeoff. That can coexist with the right solution, but it cannot replace safe caching of stable evidence: the application must still cache authorized stable retrieval results with version and TTL validation.
D:The design concern is real but different: concurrent evidence retrieval. It does not impose the specific requirement for safe caching of stable evidence; a suitable solution must cache authorized stable retrieval results with version and TTL validation and demonstrate the result.
Question 144
A retrieval-heavy response misses its latency target after adding an expensive reranker to every easy query. Which next step is most defensible?
- Adopt separate fast and deep-service paths with explicit deadlines and fallback behavior.
- Separate rarely changing descriptive text from authoritative real-time price retrieval.
- Run accuracy-versus-latency evaluations across representative hard and easy queries.
- Apply reranking selectively to ambiguous or low-confidence retrievals.
Explanation
Correct Answer
D:Quality-improving stages should be gated by measured incremental benefit rather than universally applied. Measure adaptive retrieval quality cost under representative concurrency and source freshness constraints; low latency is not success when evidence for adaptive retrieval quality cost is missing.
Incorrect Answers
A:It emphasizes quality-aware deadline management without closing the stated gap in adaptive retrieval quality cost. By contrast, the remedy to apply reranking selectively to ambiguous or low-confidence retrievals acts on the condition that makes this scenario fail.
B:That is primarily a technique for field-specific freshness guarantees. It lacks the explicit measure for adaptive retrieval quality cost that is necessary here, namely a system that will apply reranking selectively to ambiguous or low-confidence retrievals.
C:Improving validated speed and recall tradeoff does not imply adaptive retrieval quality cost. The selected design must apply reranking selectively to ambiguous or low-confidence retrievals, whereas this choice omits that direct protection or verification step.
Question 145
An integration retries all failed API calls immediately during provider throttling and makes outages worse. Which approach should be recommended?
- Use exponential backoff with jitter and a bounded retry budget for transient failures.
- Place integration gateways near supported workloads while respecting the data residency boundary.
- Adopt separate fast and deep-service paths with explicit deadlines and fallback behavior.
- Run independent reads concurrently and merge results with bounded timeouts.
Explanation
Correct Answer
A:Coordinated immediate retries amplify load; paced retries preserve service stability. Measure throttling-aware retry policy under representative concurrency and source freshness constraints; low latency is not success when evidence for throttling-aware retry policy is missing.
Incorrect Answers
B:The control would be justified if locality-aware integration topology were the limiting factor. In this scenario the decisive target is throttling-aware retry policy; success requires an architecture that will use exponential backoff with jitter and a bounded retry budget for transient failures.
C:The option belongs to the broader work on quality-aware deadline management, but it cannot by itself deliver throttling-aware retry policy. A decision on this case needs the team to use exponential backoff with jitter and a bounded retry budget for transient failures and verify that outcome.
D:This shifts attention toward concurrent evidence retrieval when the relevant decision is about throttling-aware retry policy. Unless the architecture can use exponential backoff with jitter and a bounded retry budget for transient failures, the failure described in the stem remains unresolved.
Question 146
A document assistant gives incomplete answers because a tight timeout truncates a mandatory evidence lookup. What is the best production design response?
- Run accuracy-versus-latency evaluations across representative hard and easy queries.
- Run independent reads concurrently and merge results with bounded timeouts.
- Adopt separate fast and deep-service paths with explicit deadlines and fallback behavior.
- Place integration gateways near supported workloads while respecting the data residency boundary.
Explanation
Correct Answer
C:Meeting latency objectives should not silently omit sources required for high-stakes answers. Measure quality-aware deadline management under representative concurrency and source freshness constraints; low latency is not success when evidence for quality-aware deadline management is missing.
Incorrect Answers
A:The design concern is real but different: validated speed and recall tradeoff. It does not impose the specific requirement for quality-aware deadline management; a suitable solution must adopt separate fast and deep-service paths with explicit deadlines and fallback behavior and demonstrate the result.
B:This addresses concurrent evidence retrieval; however, the problem asks for quality-aware deadline management. Without the safeguard to adopt separate fast and deep-service paths with explicit deadlines and fallback behavior, the original failure mode is still possible.
D:There is a legitimate use case for locality-aware integration topology, but it does not supply the test or enforcement needed for quality-aware deadline management. The design needs to adopt separate fast and deep-service paths with explicit deadlines and fallback behavior.
Question 147
A multi-region client routes every request through a distant integration service despite local approved endpoints. Which change most directly satisfies this requirement?
- Run accuracy-versus-latency evaluations across representative hard and easy queries.
- Place integration gateways near supported workloads while respecting the data residency boundary.
- Trigger high-cost tool calls only when the task needs current account data.
- Use exponential backoff with jitter and a bounded retry budget for transient failures.
- Apply reranking selectively to ambiguous or low-confidence retrievals.
Explanation
Correct Answer
B:Network distance and cross-region transit are part of end-to-end response latency. Measure locality-aware integration topology under representative concurrency and source freshness constraints; low latency is not success when evidence for locality-aware integration topology is missing.
Incorrect Answers
A:This recommendation targets validated speed and recall tradeoff at a different point in the workflow. It leaves locality-aware integration topology outside the design contract and does not require the team to place integration gateways near supported workloads while respecting the data residency boundary.
C:This would make sense as an additional project activity for relevance-driven tool execution. The priority required here is locality-aware integration topology, so the workflow still needs a control that will place integration gateways near supported workloads while respecting the data residency boundary.
D:By itself, this improves throttling-aware retry policy, but the case is not asking the architect to optimize that dimension. The requirement is locality-aware integration topology, supported by an approach to place integration gateways near supported workloads while respecting the data residency boundary.
E:Its main effect concerns adaptive retrieval quality cost. The case explicitly hinges on locality-aware integration topology; to resolve it reliably, the implementation has to place integration gateways near supported workloads while respecting the data residency boundary, which this choice does not establish.
Question 148
An application prefetches dozens of tools before knowing whether the user request is simple. What design decision would best fit these facts?
- Place integration gateways near supported workloads while respecting the data residency boundary.
- Use lazy tool discovery and call only capabilities required for the resolved intent.
- Use exponential backoff with jitter and a bounded retry budget for transient failures.
- Trigger high-cost tool calls only when the task needs current account data.
Explanation
Correct Answer
B:Deferred initialization reduces unnecessary context and connection overhead for common requests. Measure on-demand capability loading under representative concurrency and source freshness constraints; low latency is not success when evidence for on-demand capability loading is missing.
Incorrect Answers
A:This shifts attention toward locality-aware integration topology when the relevant decision is about on-demand capability loading. Unless the architecture can use lazy tool discovery and call only capabilities required for the resolved intent, the failure described in the stem remains unresolved.
C:Although throttling-aware retry policy can be valuable, it is not the acceptance condition in the question. The missing provision is on-demand capability loading, addressed by the control to use lazy tool discovery and call only capabilities required for the resolved intent.
D:It is aimed at a different outcome, relevance-driven tool execution. No evidence is given that it provides on-demand capability loading; the needed mechanism is to use lazy tool discovery and call only capabilities required for the resolved intent at the appropriate boundary.
Question 149
A quality regression follows a speed optimization that halves retrieval depth. Which intervention most directly addresses the failure?
- Apply reranking selectively to ambiguous or low-confidence retrievals.
- Cache authorized stable retrieval results with version and TTL validation.
- Run accuracy-versus-latency evaluations across representative hard and easy queries.
- Run independent reads concurrently and merge results with bounded timeouts.
Explanation
Correct Answer
C:Latency gains are not acceptable if they disproportionately destroy evidence recall on critical cases. Measure validated speed and recall tradeoff under representative concurrency and source freshness constraints; low latency is not success when evidence for validated speed and recall tradeoff is missing.
Incorrect Answers
A:There is a legitimate use case for adaptive retrieval quality cost, but it does not supply the test or enforcement needed for validated speed and recall tradeoff. The design needs to run accuracy-versus-latency evaluations across representative hard and easy queries.
B:Consider its actual operating scope: safe caching of stable evidence. The requested assurance is validated speed and recall tradeoff, which calls for an implementation that can run accuracy-versus-latency evaluations across representative hard and easy queries, not merely for a related improvement.
D:The proposed measure is aimed at concurrent evidence retrieval, not validated speed and recall tradeoff. The architecture must instead run accuracy-versus-latency evaluations across representative hard and easy queries; a separate change is still needed for validated speed and recall tradeoff.
Question 150
A production agent times out but logs only the final error, hiding which MCP call stalled. What should the solution architect prioritize?
- Capture tool-call counts, token usage, costs and task completion per trace.
- Record spans for model calls, tool dispatch, connector execution and result ingestion.
- Break down failure metrics by connector version, tenant, scope and response code.
- Log selected tool names, arguments and task categories for audited error analysis.
Explanation
Correct Answer
B:Distributed traces reveal the slow dependency and its contribution to the overall deadline. Correlate tool-level distributed tracing with trace IDs, component versions and tool outcomes so responders can distinguish model errors from connector failures.
Incorrect Answers
A:By itself, this improves per-task agent cost attribution, but the case is not asking the architect to optimize that dimension. The requirement is tool-level distributed tracing, supported by an approach to record spans for model calls, tool dispatch, connector execution and result ingestion.
C:The option might improve segmented authorization telemetry; that remains insufficient for tool-level distributed tracing. The case is resolved by tool-level distributed tracing, which depends on a design that will record spans for model calls, tool dispatch, connector execution and result ingestion.
D:The proposed action prioritizes tool-selection diagnostic evidence. That can coexist with the right solution, but it cannot replace tool-level distributed tracing: the application must still record spans for model calls, tool dispatch, connector execution and result ingestion.