FDE PRACTICE EXAM

FDE 知识体系模拟考试第二套（英文）

本练习使用 TIIDA FDE Body of Knowledge Sample Examination Questions Set 12 英文题库，包含 20 道单选题、20 道多选题和 4 道论述题。客观题自动判分，论述题提交后显示参考作答范围与评分要点，供学习者自评。

88 道题 120 分自动判定 80 分论述自评 200 分总分

一、单项选择题

每题只有 1 个最佳答案，每题 1 分。

1 1 分

[FDE-01 | Ch. 5–6] A public transit operator asks for an AI assistant to “reduce service complaints.” Interviews reveal that most complaints are triggered by inconsistent disruption notices rather than slow agent response. What should the FDE do first?

A Map the disruption-communication workflow, ownership, evidence sources, and measurable complaint drivers before choosing an AI intervention. B Prototype a chatbot first so stakeholders can react to something concrete. The team could then use that choice as the default for implementation planning and reporting. C Optimize response-generation latency because complaint volume is high. This approach would also reduce the number of design decisions that need to be revisited later. D Adopt the communications team’s preferred solution and validate after launch. This would also standardize the initial operating approach across teams and locations.

1 1 分

2 1 分

[FDE-01 | Ch. 7–8] A sponsor says a field-service application must work “offline.” Which requirement is most actionable?

A Require every feature to work identically without connectivity. The team could then use that choice as the default for implementation planning and reporting. B Define which tasks must continue offline, how long disconnection may last, what data can be cached, and how conflicts are reconciled. C Store the full production database on every device. D Treat offline capability as a design preference rather than a measurable requirement. This would also standardize the initial operating approach across teams and locations.

2 1 分

[FDE-01 | Ch. 7–8] A sponsor says a field-service application must work “offline.” Which requirement is most actionable?

3 1 分

[FDE-01 | Ch. 9–10] A pilot reduces manual review time by 30%, but error-related rework rises slightly. What is the strongest value interpretation?

A The pilot succeeded because labor time fell. B The pilot failed because any rework increase is unacceptable. C Value should be assessed against the combined outcome hypothesis, including speed, quality, downstream rework, and guardrails. D The rework metric should be excluded because it complicates the business case. This would also standardize the initial operating approach across teams and locations.

3 1 分

[FDE-01 | Ch. 9–10] A pilot reduces manual review time by 30%, but error-related rework rises slightly. What is the strongest value interpretation?

4 1 分

[FDE-02 | Ch. 11] Two subsidiaries assign different identifiers to the same supplier, and one identifier can be reused after legal-entity changes. What is the best data-foundation response?

A Use the numerically largest identifier as the enterprise key. B Keep both identifiers but do not reconcile them. C Replace all source identifiers with supplier names. D Model a durable enterprise supplier identity with source-key mappings and effective-dated legal-entity relationships.

4 1 分

[FDE-02 | Ch. 11] Two subsidiaries assign different identifiers to the same supplier, and one identifier can be reused after legal-entity changes. What is the best data-foundation response?

5 1 分

[FDE-02 | Ch. 12] A bank uses “customer relationship” to mean account ownership in one system and beneficial ownership in another. What should the enterprise model do?

A Represent the relationships as distinct governed concepts with explicit semantics and applicability. B Collapse both meanings into a single relationship to simplify reporting. The team could then use that choice as the default for implementation planning and reporting. C Rename the two columns but leave downstream logic unchanged. This approach would also reduce the number of design decisions that need to be revisited later. D Choose the definition used by the system with the largest number of records. This would also standardize the initial operating approach across teams and locations.

5 1 分

[FDE-02 | Ch. 12] A bank uses “customer relationship” to mean account ownership in one system and beneficial ownership in another. What should the enterprise model do?

6 1 分

[FDE-02 | Ch. 13–14] A CDC pipeline replays a window of changes after recovery. Some updates are re-applied and downstream aggregates drift. Which control is most important?

A Increase batch size so replay finishes faster. The team could then use that choice as the default for implementation planning and reporting. B Use idempotent change application with stable event/version identity and reconciliation checks. C Disable replay and rely on manual correction. D Keep only the latest source snapshot and discard the event history. This would also standardize the initial operating approach across teams and locations.

6 1 分

[FDE-02 | Ch. 13–14] A CDC pipeline replays a window of changes after recovery. Some updates are re-applied and downstream aggregates drift. Which control is most important?

7 1 分

[FDE-02 | Ch. 15–16] A regulated analytics platform meets uptime targets, but sensitive data is copied into temporary analyst tables that are never cleaned up. What is the best conclusion?

A The platform is compliant because access is authenticated. B The issue is operational, not governance-related. C Production readiness must include lifecycle controls for sensitive copies, retention, least privilege, and auditable cleanup. D Temporary tables are exempt from governance because they are not systems of record. This would also standardize the initial operating approach across teams and locations.

7 1 分

[FDE-02 | Ch. 15–16] A regulated analytics platform meets uptime targets, but sensitive data is copied into temporary analyst tables that are never cleaned up. What is the best conclusion?

8 1 分

[FDE-03 | Ch. 17] A pricing workflow uses a fixed formula, approved thresholds, and authoritative cost data. Users ask for an LLM because they want a “smarter” solution. What is the best starting point?

A Use an agent so pricing rules can evolve at runtime. B Use RAG to calculate the formula from policy documents. C Use an LLM and compare results with the rule later. D Implement deterministic logic first and reserve AI for genuinely ambiguous supporting tasks.

8 1 分

9 1 分

[FDE-03 | Ch. 18] An LLM drafts audit summaries, but reviewers need every claim linked to a source record and unsupported claims flagged. What design best addresses this?

A Require structured claim-evidence outputs, validate source references, and evaluate unsupported-claim behavior. B Increase temperature so the model considers more interpretations. This would also standardize the initial operating approach across teams and locations. C Use a larger context window without changing the output contract. The team could then use that choice as the default for implementation planning and reporting. D Ask reviewers to infer which statements came from the evidence. This approach would also reduce the number of design decisions that need to be revisited later.

9 1 分

[FDE-03 | Ch. 18] An LLM drafts audit summaries, but reviewers need every claim linked to a source record and unsupported claims flagged. What design best addresses this?

10 1 分

[FDE-03 | Ch. 19] An agent can schedule maintenance and also cancel existing work orders. What control most directly limits harmful side effects?

A Keep all action tools available but add a warning in the system prompt. The team could then use that choice as the default for implementation planning and reporting. B Enforce tool-specific permissions, parameter validation, approval thresholds, and auditable action scope outside the model. C Let the agent infer which cancellations are sensitive from historical examples. This would also standardize the initial operating approach across teams and locations. D Use a single generic “execute” tool so orchestration is simpler.

10 1 分

[FDE-03 | Ch. 19] An agent can schedule maintenance and also cancel existing work orders. What control most directly limits harmful side effects?

11 1 分

[FDE-03 | Ch. 20–22] A retrieval assistant performs well overall but fails disproportionately when two valid policies conflict by region. What should the FDE do?

A Increase top-k until both policies always appear. B Exclude conflicted cases from the benchmark. C Define authority/applicability rules, test conflict resolution by segment, and monitor production escalation behavior. D Average the wording of both policies into one answer. This would also standardize the initial operating approach across teams and locations.

11 1 分

[FDE-03 | Ch. 20–22] A retrieval assistant performs well overall but fails disproportionately when two valid policies conflict by region. What should the FDE do?

12 1 分

[FDE-04 | Ch. 23] A solution reads from several systems and writes decisions back to one of them, but the architecture diagram does not show who owns the final business state. What is the most important correction?

A Add more infrastructure detail to the diagram. B Document the programming language used by each service. C Move all data into a single database. D Make systems of record, state ownership, decision authority, and side-effect boundaries explicit.

12 1 分

13 1 分

[FDE-04 | Ch. 24] A case-management workflow may wait several hours for an external regulator response. Which architecture is most appropriate?

A Persist workflow state durably and resume from explicit transitions when the external response arrives. B Keep the user session open until the regulator responds. This would also standardize the initial operating approach across teams and locations. C Store the entire process only in model context. This approach would also reduce the number of design decisions that need to be revisited later. D Restart the process from step one after every timeout. The team could then use that choice as the default for implementation planning and reporting.

13 1 分

[FDE-04 | Ch. 24] A case-management workflow may wait several hours for an external regulator response. Which architecture is most appropriate?

14 1 分

[FDE-04 | Ch. 25–26] A team is unsure whether warehouse supervisors will trust an AI-generated exception ranking. What is the best MVS experiment?

A Build full production automation so trust can be measured at scale. The team could then use that choice as the default for implementation planning and reporting. B Run a bounded advisory-mode trial with representative exceptions, explanations, supervisor overrides, and predefined trust/usefulness measures. C Interview only senior executives because they approve the project. D Measure model accuracy offline and infer that operational trust will follow. This would also standardize the initial operating approach across teams and locations.

14 1 分

[FDE-04 | Ch. 25–26] A team is unsure whether warehouse supervisors will trust an AI-generated exception ranking. What is the best MVS experiment?

15 1 分

[FDE-04 | Ch. 27–28] A trial beats the baseline only after analysts remove a difficult site that was included in the original plan. What should the FDE recommend?

A Report the filtered result because it reflects realistic conditions. B Change the original inclusion criteria so the analysis remains consistent. This would also standardize the initial operating approach across teams and locations. C Preserve the original result, analyze the site as a segment, and document any follow-up experiment separately. D Stop all experimentation because the first trial was mixed.

15 1 分

[FDE-04 | Ch. 27–28] A trial beats the baseline only after analysts remove a difficult site that was included in the original plan. What should the FDE recommend?

16 1 分

[FDE-05 | Ch. 29] A service passes functional tests but the team has not tested credential expiry, dependency outage, or rollback. Is it production-ready?

A Yes, because application logic is correct. B Yes, if the business sponsor accepts the risk. C Only if the launch audience is small. D No; production readiness also requires failure behavior, recovery, observability, access controls, and support ownership.

16 1 分

[FDE-05 | Ch. 29] A service passes functional tests but the team has not tested credential expiry, dependency outage, or rollback. Is it production-ready?

17 1 分

[FDE-05 | Ch. 30–31] A new API version changes an enum value used by an external partner. What is the safest release approach?

A Validate contract compatibility, version behavior, monitoring, and rollback before progressive rollout. B Deploy immediately and fix partner failures afterward. The team could then use that choice as the default for implementation planning and reporting. C Remove schema validation to avoid rejecting changed values. This would also standardize the initial operating approach across teams and locations. D Freeze all future API evolution permanently. This approach would also reduce the number of design decisions that need to be revisited later.

17 1 分

[FDE-05 | Ch. 30–31] A new API version changes an enum value used by an external partner. What is the safest release approach?

18 1 分

[FDE-05 | Ch. 32–34] A copilot reaches 90% weekly usage, yet business outcomes remain flat. What should the FDE do next?

A Treat usage as proof of value because adoption is strong. This would also standardize the initial operating approach across teams and locations. B Investigate workflow behavior, output quality, displacement effects, and outcome metrics against the original baseline. C Increase the usage target before reviewing outcomes. The team could then use that choice as the default for implementation planning and reporting. D Stop measuring value until users have had more time.

18 1 分

[FDE-05 | Ch. 32–34] A copilot reaches 90% weekly usage, yet business outcomes remain flat. What should the FDE do next?

19 1 分

[FDE-06 | Ch. 35–37] During a workshop, an FDE is pressed to give a definitive cybersecurity interpretation outside their expertise. What is the best response?

A Offer a cautious opinion so the meeting can move forward. B Decline all further work until a security expert joins. C State the boundary, capture the dependency, involve the proper authority, and continue work that does not depend on the unresolved point. D Ask the customer to accept responsibility for any incorrect interpretation. This would also standardize the initial operating approach across teams and locations.

19 1 分

[FDE-06 | Ch. 35–37] During a workshop, an FDE is pressed to give a definitive cybersecurity interpretation outside their expertise. What is the best response?

20 1 分

[FDE-06 | Ch. 38–40] Which behavior best demonstrates mature professional mastery after a near miss?

A Keep the event informal to avoid creating unnecessary scrutiny. B Focus the review on the individual who made the last mistake. C Patch the immediate defect and move on quickly. D Examine technical and decision conditions, record reusable lessons, and ensure corrective actions change future practice.

20 1 分

[FDE-06 | Ch. 38–40] Which behavior best demonstrates mature professional mastery after a near miss?

二、多项选择题

每题有 2–4 个正确答案，全部选对且未选错才得 2 分。

21 2 分

[FDE-01 | Ch. 5–6] Which evidence sources are useful when discovering why a claims process is slow? Select all that apply.

A Process timestamps and exception data. B Observation of actual handler workflows. C Interviews with handlers, supervisors, and downstream teams. D Only the executive sponsor’s preferred solution.

21 2 分

[FDE-01 | Ch. 5–6] Which evidence sources are useful when discovering why a claims process is slow? Select all that apply.

A Process timestamps and exception data. B Observation of actual handler workflows. C Interviews with handlers, supervisors, and downstream teams. D Only the executive sponsor’s preferred solution.

22 2 分

[FDE-01 | Ch. 7–8] Which practices strengthen a well-framed requirement? Select all that apply.

A Prescribe the technology before alternatives are evaluated. B Define measurable outcomes and constraints. C Distinguish assumptions from verified facts. D Clarify decision rights and non-goals.

22 2 分

[FDE-01 | Ch. 7–8] Which practices strengthen a well-framed requirement? Select all that apply.

A Prescribe the technology before alternatives are evaluated. B Define measurable outcomes and constraints. C Distinguish assumptions from verified facts. D Clarify decision rights and non-goals.

23 2 分

[FDE-01 | Ch. 9–10] Which elements belong in a credible value hypothesis? Select all that apply.

A A baseline or reference condition. B A guaranteed financial return before testing. C Outcome measures connected to the workflow. D Assume a successful demo eliminates the need for operational validation.

23 2 分

[FDE-01 | Ch. 9–10] Which elements belong in a credible value hypothesis? Select all that apply.

24 2 分

[FDE-02 | Ch. 11–12] Which elements matter when designing an enterprise ontology or semantic layer? Select all that apply.

A A rule that identical column names imply identical meaning. B Entity identity and relationships. C Governed definitions and provenance. D Applicability rules for context or time.

24 2 分

[FDE-02 | Ch. 11–12] Which elements matter when designing an enterprise ontology or semantic layer? Select all that apply.

A A rule that identical column names imply identical meaning. B Entity identity and relationships. C Governed definitions and provenance. D Applicability rules for context or time.

25 2 分

[FDE-02 | Ch. 13] Which controls improve reliable ingestion from a flaky external API? Select all that apply.

A Bounded retries with backoff. B Silent dropping of malformed responses. C Checkpointing or durable progress. D Rate-limit awareness.

25 2 分

[FDE-02 | Ch. 13] Which controls improve reliable ingestion from a flaky external API? Select all that apply.

A Bounded retries with backoff. B Silent dropping of malformed responses. C Checkpointing or durable progress. D Rate-limit awareness.

26 2 分

[FDE-02 | Ch. 14–15] Which controls strengthen a sensitive analytics pipeline? Select all that apply.

A Default export rights for all analysts. B Least-privilege access. C Assume a successful demo eliminates the need for operational validation. D Lineage for material transformations.

26 2 分

[FDE-02 | Ch. 14–15] Which controls strengthen a sensitive analytics pipeline? Select all that apply.

A Default export rights for all analysts. B Least-privilege access. C Assume a successful demo eliminates the need for operational validation. D Lineage for material transformations.

27 2 分

[FDE-02 | Ch. 16] Which production signals are useful for operating a streaming data product? Select all that apply.

A Freshness or lag. B Only whether the orchestration job is marked successful. C Failure/retry rate. D Backlog or partition skew.

27 2 分

[FDE-02 | Ch. 16] Which production signals are useful for operating a streaming data product? Select all that apply.

A Freshness or lag. B Only whether the orchestration job is marked successful. C Failure/retry rate. D Backlog or partition skew.

28 2 分

[FDE-03 | Ch. 17–18] Which conditions support using generative AI for an enterprise task? Select all that apply.

A The task requires interpretation or generation of unstructured language. B Probabilistic errors can be bounded or reviewed. C A deterministic method cannot fully satisfy the need. D The organization wants AI regardless of task fit.

28 2 分

[FDE-03 | Ch. 17–18] Which conditions support using generative AI for an enterprise task? Select all that apply.

29 2 分

[FDE-03 | Ch. 19] Which controls are appropriate for an enterprise agent with write access? Select all that apply.

A Model-decided credentials stored in prompts. B Explicit tool schemas. C Bounded permissions. D Idempotency or duplicate-action protection.

29 2 分

[FDE-03 | Ch. 19] Which controls are appropriate for an enterprise agent with write access? Select all that apply.

A Model-decided credentials stored in prompts. B Explicit tool schemas. C Bounded permissions. D Idempotency or duplicate-action protection.

30 2 分

[FDE-03 | Ch. 20] Which practices strengthen enterprise retrieval and grounding? Select all that apply.

A Track source provenance. B Treat semantic similarity as proof of truth. C Apply authority and freshness rules. D Assume a successful demo eliminates the need for operational validation.

30 2 分

[FDE-03 | Ch. 20] Which practices strengthen enterprise retrieval and grounding? Select all that apply.

A Track source provenance. B Treat semantic similarity as proof of truth. C Apply authority and freshness rules. D Assume a successful demo eliminates the need for operational validation.

31 2 分

[FDE-03 | Ch. 21–22] Which evaluation practices are appropriate before broad release? Select all that apply.

A Representative and difficult cases. B Segment-level metrics for high-risk groups. C Failure, abstention, and guardrail testing. D Monitor production drift and workflow outcome signals.

31 2 分

[FDE-03 | Ch. 21–22] Which evaluation practices are appropriate before broad release? Select all that apply.

A Representative and difficult cases. B Segment-level metrics for high-risk groups. C Failure, abstention, and guardrail testing. D Monitor production drift and workflow outcome signals.

32 2 分

[FDE-04 | Ch. 23–24] Which questions belong in an end-to-end architecture review? Select all that apply.

A Which framework has the largest developer community? B Where does authoritative state live? C Which components can create side effects? D How do failures propagate and recover?

32 2 分

[FDE-04 | Ch. 23–24] Which questions belong in an end-to-end architecture review? Select all that apply.

A Which framework has the largest developer community? B Where does authoritative state live? C Which components can create side effects? D How do failures propagate and recover?

33 2 分

[FDE-04 | Ch. 25] Which practices are consistent with evidence-driven prototyping? Select all that apply.

A Identify the uncertainty being reduced. B Treat prototype code as production-ready by default. C Assume a successful demo eliminates the need for operational validation. D Define what evidence would change the decision.

33 2 分

[FDE-04 | Ch. 25] Which practices are consistent with evidence-driven prototyping? Select all that apply.

34 2 分

[FDE-04 | Ch. 26] Which issues can make a promising prototype infeasible for production? Select all that apply.

A Unacceptable latency or cost at expected scale. B Unavailable authoritative data. C Security or regulatory constraints. D The prototype uses fewer technologies than planned.

34 2 分

[FDE-04 | Ch. 26] Which issues can make a promising prototype infeasible for production? Select all that apply.

A Unacceptable latency or cost at expected scale. B Unavailable authoritative data. C Security or regulatory constraints. D The prototype uses fewer technologies than planned.

35 2 分

[FDE-04 | Ch. 27–28] Which practices strengthen experimental validity and decision records? Select all that apply.

A Predefine success criteria where practical. B Rewrite the original hypothesis after seeing results. C Record material assumptions and versions. D Include failure cases and limitations.

35 2 分

[FDE-04 | Ch. 27–28] Which practices strengthen experimental validity and decision records? Select all that apply.

A Predefine success criteria where practical. B Rewrite the original hypothesis after seeing results. C Record material assumptions and versions. D Include failure cases and limitations.

36 2 分

[FDE-05 | Ch. 29–30] Which controls support a safer production release? Select all that apply.

A Unversioned configuration changes. B Progressive exposure. C Rollback or disable capability. D Post-release verification.

36 2 分

[FDE-05 | Ch. 29–30] Which controls support a safer production release? Select all that apply.

A Unversioned configuration changes. B Progressive exposure. C Rollback or disable capability. D Post-release verification.

37 2 分

[FDE-05 | Ch. 31–32] Which conditions can create adoption risk even when integration works technically? Select all that apply.

A The workflow adds hidden verification work. B Roles or decision rights become unclear. C The API returns HTTP 200. D Incentives conflict with the new behavior.

37 2 分

[FDE-05 | Ch. 31–32] Which conditions can create adoption risk even when integration works technically? Select all that apply.

A The workflow adds hidden verification work. B Roles or decision rights become unclear. C The API returns HTTP 200. D Incentives conflict with the new behavior.

38 2 分

[FDE-05 | Ch. 33–34] Which evidence supports scaling a successful pilot? Select all that apply.

A Value persists beyond the pilot team. B Every site must use exactly the same workflow. C Support burden remains acceptable. D Assume a successful demo eliminates the need for operational validation.

38 2 分

[FDE-05 | Ch. 33–34] Which evidence supports scaling a successful pilot? Select all that apply.

39 2 分

[FDE-06 | Ch. 35–37] Which behaviors demonstrate strong FDE judgment under ambiguity? Select all that apply.

A Equate customer empathy with agreeing to every request. B State uncertainty rather than invent facts. C Make reversible decisions when evidence is incomplete. D Escalate when authority or specialist competence is required.

39 2 分

[FDE-06 | Ch. 35–37] Which behaviors demonstrate strong FDE judgment under ambiguity? Select all that apply.

40 2 分

[FDE-06 | Ch. 38–40] Which behaviors support long-term professional mastery? Select all that apply.

A Reflect on failures and near misses. B Create reusable patterns with known limits. C Mentor others in reasoning as well as tools. D Use peer review to improve consequential decisions.

40 2 分

[FDE-06 | Ch. 38–40] Which behaviors support long-term professional mastery? Select all that apply.

A Reflect on failures and near misses. B Create reusable patterns with known limits. C Mentor others in reasoning as well as tools. D Use peer review to improve consequential decisions.

三、论述题

请结合场景作答。提交后可查看参考答案与评分要点，论述题采用自评方式。

41 10 分

[FDE-02 | Ch. 11–16] Business Scenario
A global renewable-energy company is building a unified asset-performance data product across wind farms, solar sites, and battery facilities. Data arrives from SCADA systems, vendor portals, maintenance work orders, weather feeds, finance systems, and technician spreadsheets. Asset identifiers change after component replacements; timestamps use mixed local/UTC conventions; vendors resend historical telemetry after connectivity outages; maintenance teams correct failure classifications weeks later; and some free-text notes contain employee or landowner information. Operations wants near-real-time anomaly visibility, finance wants monthly loss attribution, and data science wants point-in-time correct features for predictive maintenance.
Candidate Task: Design the enterprise data solution. Address source authority and identity, ontology/semantic modeling, event time and late corrections, integration patterns, quality/lineage/observability, sensitive-data governance, point-in-time correctness, recovery/replay, scalability, operating ownership, and evidence required before the data product is trusted for analytics or AI.

您的回答

参考答案与自评要点

Design the enterprise data solution.
Source authority and identity.
Ontology/semantic modeling.
Event time and late corrections.
Integration patterns.
Quality/lineage/observability.
Sensitive-data governance.
Point-in-time correctness.
Recovery/replay.
Scalability.
Operating ownership.
Evidence required before the data product is trusted for analytics or AI.

以下为练习自评，不代表 TIIDA 官方考试评分。

Scenario understanding and problem framing（2 分） Identifies business goals, constraints, stakeholders, risks, and unknowns. Solution design and technical correctness（3 分） Proposes a coherent design that addresses the major capabilities required by the scenario. Tradeoffs, governance, and decision boundaries（2 分） Explains material risks, authority boundaries, controls, and alternatives. Validation and measurable outcomes（2 分） Defines evidence, tests, metrics, readiness gates, and failure criteria. Professional judgment and communication（1 分） Makes assumptions explicit, prioritizes decisions, and communicates a defensible recommendation.

41 10 分

您的回答

参考答案与自评要点

Design the enterprise data solution.
Source authority and identity.
Ontology/semantic modeling.
Event time and late corrections.
Integration patterns.
Quality/lineage/observability.
Sensitive-data governance.
Point-in-time correctness.
Recovery/replay.
Scalability.
Operating ownership.
Evidence required before the data product is trusted for analytics or AI.

以下为练习自评，不代表 TIIDA 官方考试评分。

42 10 分

[FDE-03 | Ch. 17–22] Business Scenario
A multinational procurement organization wants a sourcing copilot that reads contracts, supplier policies, purchase history, risk reports, and negotiation emails. It may draft negotiation positions, retrieve approved clauses, compare bids, and create approval requests. Some contracts have amendments with overlapping effective dates; regional policies differ; external supplier documents may contain malicious instructions; discount and commitment thresholds vary by user authority; and executives want the system to remember supplier preferences across sessions without retaining unnecessary personal data.
Candidate Task: Design a bounded AI/agent architecture. Cover AI-versus-deterministic task selection, retrieval authority and applicability, prompt-injection defenses, tool schemas and authorization, memory boundaries, human approval, evaluation/red-team design, auditability, production monitoring, failure fallback, and the progression from advisory mode to any limited autonomy.

您的回答

参考答案与自评要点

Design a bounded AI/agent architecture.
AI-versus-deterministic task selection.
Retrieval authority and applicability.
Prompt-injection defenses.
Tool schemas and authorization.
Memory boundaries.
Human approval.
Evaluation/red-team design.
Auditability.
Production monitoring.
Failure fallback.
The progression from advisory mode to any limited autonomy.

以下为练习自评，不代表 TIIDA 官方考试评分。

42 10 分

您的回答

参考答案与自评要点

Design a bounded AI/agent architecture.
AI-versus-deterministic task selection.
Retrieval authority and applicability.
Prompt-injection defenses.
Tool schemas and authorization.
Memory boundaries.
Human approval.
Evaluation/red-team design.
Auditability.
Production monitoring.
Failure fallback.
The progression from advisory mode to any limited autonomy.

以下为练习自评，不代表 TIIDA 官方考试评分。

43 10 分

[FDE-04 | Ch. 23–28] Business Scenario
A regional hospital network wants to reduce operating-room schedule disruption using a recommendation service that predicts procedure overruns and proposes schedule adjustments. The prototype was trained on historical cases, but some hospitals record procedure stages differently, surgeon preferences are not consistently structured, urgent cases can override normal sequencing, and staffing/bed constraints sit in separate systems. Leadership wants an enterprise rollout after a strong retrospective accuracy result.
Candidate Task: Provide an end-to-end architecture and validation recommendation. Address systems of record and authority, integration boundaries, the minimum viable solution, feasibility risks, human decision points, retrospective leakage and prospective validation, baselines and outcome measures, failure/degraded modes, architecture fitness criteria, and the decision records needed before rollout.

您的回答

参考答案与自评要点

Provide an end-to-end architecture and validation recommendation.
Systems of record and authority.
Integration boundaries.
The minimum viable solution.
Feasibility risks.
Human decision points.
Retrospective leakage and prospective validation.
Baselines and outcome measures.
Failure/degraded modes.
Architecture fitness criteria.
The decision records needed before rollout.

以下为练习自评，不代表 TIIDA 官方考试评分。

43 10 分

您的回答

参考答案与自评要点

Provide an end-to-end architecture and validation recommendation.
Systems of record and authority.
Integration boundaries.
The minimum viable solution.
Feasibility risks.
Human decision points.
Retrospective leakage and prospective validation.
Baselines and outcome measures.
Failure/degraded modes.
Architecture fitness criteria.
The decision records needed before rollout.

以下为练习自评，不代表 TIIDA 官方考试评分。

44 10 分

[FDE-05 | Ch. 29–34] Business Scenario
A logistics company has successfully piloted a dispatch assistant in two distribution centers. It recommends dock assignments and departure sequencing based on inbound loads, labor, yard conditions, and delivery commitments. The company wants to expand to 70 sites. Sites run different versions of the yard-management system, network reliability varies, local operating procedures differ, and the pilot team still handles incidents informally. Recommendation models and rules change weekly. Some supervisors follow suggestions closely while others ignore them, and leadership wants evidence that the system improves on-time departure rather than merely increasing usage.
Candidate Task: Develop the production deployment and scaling plan. Cover readiness gates, contract/version compatibility, progressive rollout, feature flags and rollback, offline/degraded operation, observability and incident response, L1/L2/L3 support, training and workflow adoption, local-versus-global configuration, outcome and guardrail metrics, and explicit criteria for pausing or continuing expansion.

您的回答

参考答案与自评要点

Develop the production deployment and scaling plan.
Readiness gates.
Contract/version compatibility.
Progressive rollout.
Feature flags and rollback.
Offline/degraded operation.
Observability and incident response.
L1/L2/L3 support.
Training and workflow adoption.
Local-versus-global configuration.
Outcome and guardrail metrics.
Explicit criteria for pausing or continuing expansion.

以下为练习自评，不代表 TIIDA 官方考试评分。

44 10 分

您的回答

参考答案与自评要点

Develop the production deployment and scaling plan.
Readiness gates.
Contract/version compatibility.
Progressive rollout.
Feature flags and rollback.
Offline/degraded operation.
Observability and incident response.
L1/L2/L3 support.
Training and workflow adoption.
Local-versus-global configuration.
Outcome and guardrail metrics.
Explicit criteria for pausing or continuing expansion.

以下为练习自评，不代表 TIIDA 官方考试评分。

完成后提交试卷

客观题将自动判分；论述题提交后按参考要点自行评估。

FDE 知识体系模拟考试 第二套（英文）

参考答案与自评要点

参考答案与自评要点

参考答案与自评要点

参考答案与自评要点

参考答案与自评要点

参考答案与自评要点

参考答案与自评要点

参考答案与自评要点

完成后提交试卷

成绩与答案解析

FDE 知识体系模拟考试第二套（英文）