Wrong-Answer Trap Library
335 traps catalogued from your 285 practice questions. Pattern recognition is the moat. Browse, filter, or quiz yourself.
This is a wrong-answer trap:
Which ECO task does this trap most likely appear in?
- D1-001 · Domain I · I.5 — Manage accountability documentation and audit trail cross-domain-link
Trap: Re-run the prediction on the same inputs and submit the resulting output as the response
Why wrong: Wrong-answer trap. Re-running is the correct procedure when validating reproducibility after a suspected inference defect. As a regulatory response it substitutes today's model version and data for the original decision record.
Correct answer was: C
Provide the logged audit trail for that decision: input, model version, output, rationale
- D1-002 · Domain I · I.3 — Conduct bias checks documentation-required
Trap: Record the disparity in the risk register, accept the residual, and monitor production
Why wrong: Wrong-answer trap. Registering a residual risk, accepting it, and monitoring it is correct once a mitigation has been chosen and applied — the remainder is what gets accepted. Done now, it books an unmitigated disparity as an accepted risk.
Correct answer was: D
Document the finding, weigh mitigation options with stakeholders, monitor in production
- D1-003 · Domain I · I.4 — Monitor regulatory and policy compliance phase-identification
Trap: Phases I, II, and VI carry it; the modeling phases are technical build work
Why wrong: Wrong-answer trap. The named phases are where responsible-AI work is most visible, which makes the carve-out easy to accept. The modeling phases are exactly where technique selection settles explainability and where training data carries bias into the model.
Correct answer was: D
Every phase carries its own responsible-AI work, so obligations are already in flight
- D1-004 · Domain I · I.2 — Manage AI/ML transparency cross-domain-link
Trap: Lock the technique and have compliance review the post-hoc explanations after deployment
Why wrong: Wrong-answer trap. Locking the technique and having compliance review the explanations after deployment is workable for a low-stakes internal model. For a high-stakes regulated decision it reverses the order transparency work is supposed to follow.
Correct answer was: A
Surface the explainability trade-off to compliance and stakeholders before technique lock
- D1-005 · Domain I · I.1 — Oversee privacy and security plan cross-domain-link
Trap: Have the data engineer run the anonymization in parallel with the scheduled privacy review
Why wrong: Wrong-answer trap. Running work in parallel with a review is a legitimate schedule-compression tactic for reversible steps; anonymizing a training set commits the pipeline before anyone has confirmed the control is adequate.
Correct answer was: D
Coordinate with privacy and legal to validate the anonymization, then update the privacy plan
- D1-006 · Domain I · I.2 — Manage AI/ML transparency documentation-required
Trap: Systemic covers why one prediction came out as it did; decision covers how it was built
Why wrong: Wrong-answer trap. This is the correct pair of dimensions with the two definitions swapped, which is the confusion the exam tests most often on this pair.
Correct answer was: D
Systemic covers how the model was built; decision covers why one prediction came out that way
- D1-007 · Domain I · I.3 — Conduct bias checks pattern-recognition
Trap: All three forms carry equal weight when fairness compliance is being evaluated
Why wrong: Wrong-answer trap. Treating all three as equally fairness-bearing sounds balanced and thorough, but conflating the three meanings of 'bias' is exactly the confusion the exam probes.
Correct answer was: C
Bias from over- or underrepresentation of population groups across the training data
- D1-008 · Domain I · I.5 — Manage accountability documentation and audit trail documentation-required
Trap: The model version that produced each prediction served in production
Why wrong: Wrong-answer trap. Version records feel like configuration management rather than audit, but they are essential: without them nobody can tell which model produced a contested decision.
Correct answer was: A
The training run's hyperparameter settings and cross-validation scores
- D1-009 · Domain I · I.4 — Monitor regulatory and policy compliance cross-domain-link
Trap: Have the data scientist add post-hoc explanations that satisfy the new explanation rule
Why wrong: Wrong-answer trap. Adding post-hoc explanations is the right build response once legal has confirmed that a post-hoc approximation clears the regulator's bar. That determination has not been made, and it is the whole question.
Correct answer was: D
Assess the requirement with legal, then bring stakeholders the options and their impact
- D1-010 · Domain I · I.1 — Oversee privacy and security plan documentation-required
Trap: User permissions, identity verification, audit signaling, disclosure language, encryption
Why wrong: Wrong-answer trap. Access management is genuinely part of the plan, but this set substitutes identity controls for the personal-data identification and regulatory mapping that anchor it.
Correct answer was: A
Personal data identified, regulations mapped, incident response, access controls, encryption
- D1-011 · Domain I · I.2 — Manage AI/ML transparency documentation-required
Trap: Disclosure means users can opt out; consent means they are told an AI decided
Why wrong: Wrong-answer trap. This is the correct pair of obligations with the definitions reversed, which catches readers who remember both terms but not which one is which.
Correct answer was: A
Disclosure means users know AI decided; consent means they can opt out to an alternative
- D1-012 · Domain I · I.3 — Conduct bias checks oversight-vs-doing
Trap: Direct the team to rebalance the training data before the next training run
Why wrong: Wrong-answer trap — rebalancing training data is a legitimate mitigation and the natural instinct for a technical lead. Issued as a PM directive it makes a fairness call that belongs to the stakeholders who set the tolerance.
Correct answer was: B
Record the measured disparity and put the mitigation options to the stakeholders who own the tolerance
- D1-013 · Domain I · I.5 — Manage accountability documentation and audit trail cross-domain-link
Trap: Treat it as an interim location and revisit storage at the operationalization gate
Why wrong: Wrong-answer trap. Deferring a storage decision to the operationalization gate is sound for development artifacts that carry no retention obligation. The audit trail's retention clock and access restriction apply from the first logged decision.
Correct answer was: A
Require controlled access and the regulatory retention period wherever the trail is stored
- D1-014 · Domain I · I.4 — Monitor regulatory and policy compliance documentation-required
Trap: The California Consumer Privacy Act, as amended by the CPRA
Why wrong: Wrong-answer trap. CCPA and CPRA are the closest analogue in structure and consumer rights, which makes them tempting, but they govern California residents rather than EU residents.
Correct answer was: A
The General Data Protection Regulation, known as GDPR
- D1-015 · Domain I · I.1 — Oversee privacy and security plan cross-domain-link
Trap: Privacy planning, regulatory monitoring, and accountability for the clinical decisions
Why wrong: Wrong-answer trap. Every element named here is correct, and the omission is the point: privacy, compliance, and accountability are the three obligations health projects talk about most, which is exactly why a missing bias check goes unnoticed.
Correct answer was: D
Privacy planning, bias checks, regulatory monitoring, and accountability for decisions
- D1-016 · Domain I · I.2 — Manage AI/ML transparency documentation-required
Trap: Take the proposal to compliance and stakeholders to decide whether disclosure applies
Why wrong: Wrong-answer trap. Routing an ambiguous case to compliance and stakeholders is the right move when it is genuinely unclear whether an interaction counts as customer-facing AI. A retail customer service chatbot is not ambiguous, and re-opening a settled obligation for deliberation is how it quietly slips.
Correct answer was: D
Reject it, since disclosure that users are interacting with AI is required from launch onward
- D1-017 · Domain I · I.3 — Conduct bias checks iteration-trigger
Trap: Direct the data scientist to apply the calibration, record it, and monitor in production
Why wrong: Wrong-answer trap. Applying per-group calibration, recording the change, and monitoring it is exactly right once legal has cleared per-group thresholds for this use case. Without that clearance it commits the project to a practice that may itself be disparate treatment.
Correct answer was: D
Treat it as a bias incident: document it, engage legal and ethics, then weigh options
- D1-018 · Domain I · I.5 — Manage accountability documentation and audit trail documentation-required
Trap: The audit trail is one component of the project's systemic transparency documentation
Why wrong: Wrong-answer trap. A nesting relationship sounds tidy and would let one artifact discharge both obligations, but the two document different things under different tasks.
Correct answer was: C
Systemic transparency shows how the model was built; the audit trail records its use
- D1-019 · Domain I · I.4 — Monitor regulatory and policy compliance iteration-trigger
Trap: Refer the regulation to legal for analysis and continue executing the current project plan
Why wrong: Wrong-answer trap. Routing it to legal is half right, and legal without project context cannot judge which deliverables, data flows, or deployment choices the rule touches — while the project keeps committing decisions underneath the analysis.
Correct answer was: C
Assess applicability with legal, then engage stakeholders on the impact to the plan
- D1-020 · Domain I · I.1 — Oversee privacy and security plan go-no-go-gate
Trap: Record a Go and complete the outstanding privacy plan sign-off during data preparation
Why wrong: Wrong-answer trap. Recording a Go and finishing the sign-off during data preparation reads as pragmatic scheduling, and it starts irreversible work on data whose lawful use nobody has confirmed.
Correct answer was: A
No-Go — iterate within Phase II until legal's sign-off on the privacy plan is documented
- D1-021 · Domain I · I.3 — Conduct bias checks documentation-required
Trap: Documenting the known bias in the model card satisfies the transparency obligation
Why wrong: Wrong-answer trap. A model card entry does satisfy an I.2 transparency obligation, which is why this reads as a complete answer. The stakeholder asked about I.3, where the standard is measured, monitored, and mitigated.
Correct answer was: B
Bias must be measured, monitored, and mitigated; documenting it alone is not enough
- D1-022 · Domain I · I.2 — Manage AI/ML transparency documentation-required
Trap: Interpretability is applied after training; XAI is a property of the model's design
Why wrong: Wrong-answer trap. This is the correct pair of concepts with the definitions swapped, which is the confusion the exam most often tests here.
Correct answer was: D
XAI explains a trained model after the fact; interpretability is built into the model
- D1-023 · Domain I · I.5 — Manage accountability documentation and audit trail documentation-required
Trap: The organization is accountable as the deploying entity, under its published AI policy
Why wrong: Wrong-answer trap. Organizational liability is real and is how enforcement actually lands, which makes this the natural answer. The framework's requirement is narrower: a named human, not an entity, is accountable for a consequential decision.
Correct answer was: B
A named human is accountable, with the oversight tier scaled to the stakes of the decision
- D1-024 · Domain I · I.4 — Monitor regulatory and policy compliance phase-identification
Trap: In every phase except data preparation, where the technical transformations occur
Why wrong: Wrong-answer trap. The continuous framing is right and the carve-out is not: data preparation is where transformation, retention, and minimization rules bite hardest, which makes it the worst phase to exempt.
Correct answer was: D
In every phase, because each one commits decisions with regulatory consequences
- D1-025 · Domain I · I.1 — Oversee privacy and security plan iteration-trigger
Trap: Coordinate with the security team to patch, verify the fix, and document the closure
Why wrong: Wrong-answer trap. Patch, verify, document is a complete and well-governed remediation for a defect with no data exposure. It closes the hole while skipping the exposure audit and the notification obligations that make this an incident.
Correct answer was: C
Run the contingency response: assess, contain, audit exposure, notify, and document
- D1-026 · Domain I · I.2 — Manage AI/ML transparency cross-domain-link
Trap: Ask legal whether any statute in the launch markets compels an AI disclosure notice
Why wrong: Wrong-answer trap. Asking legal which markets impose a statutory duty is the right question when scoping jurisdiction-specific obligations on top of the baseline. It is the wrong test for whether to disclose, because the framework requires it whether or not a statute does.
Correct answer was: A
Block deployment and take the disclosure requirement to compliance and stakeholders
- D1-027 · Domain I · I.3 — Conduct bias checks iteration-trigger
Trap: Document the disparity in the model card so that affected users are told it exists
Why wrong: Wrong-answer trap. A model card limitation is a genuine transparency measure and reads as responsible disclosure. It changes nothing about the disparity, and telling people about an unfair outcome does not make it fair.
Correct answer was: B
Take the measured disparity to ethics and legal to determine whether it is acceptable here
- D1-028 · Domain I · I.5 — Manage accountability documentation and audit trail cross-domain-link
Trap: Regulatory compliance monitoring, since governance and compliance verification coincide
Why wrong: Wrong-answer trap. Governance and compliance are used interchangeably in everyday practice, which makes this the most tempting single answer, but compliance verification is one governance activity among several.
Correct answer was: C
Every Domain I task, since governance is where each is enforced rather than planned
- D1-029 · Domain I · I.4 — Monitor regulatory and policy compliance documentation-required
Trap: Place it in the deployment plan, since training lands closest to the go-live date
Why wrong: Wrong-answer trap. Timing training close to go-live is right for tool enablement and reads as sound scheduling, but governance literacy is needed while the decisions it governs are still being made.
Correct answer was: B
An educated workforce is a governance principle, so training is tracked as governance
- D1-030 · Domain I · I.2 — Manage AI/ML transparency documentation-required
Trap: The affected user's right to be told that an AI, rather than a person, made the call
Why wrong: Wrong-answer trap. Disclosure is the first step of the contestability path, which is why it reads like the definition. Contestability is the dispute mechanism and human review that follow it.
Correct answer was: D
The affected user's right to challenge a decision, with an explanation and human review
- D2-001 · Domain II · II.1 — Identify problem to be solved oversight-vs-doing
Trap: Coordinate a feasibility review with the data scientist on support ticket data
Why wrong: Wrong-answer trap. A feasibility review with the data scientist is correct at II.2 and reads as responsible early diligence, but II.2 evaluates the feasibility of a defined problem; run first, it tests whether data exists for a question nobody has asked.
Correct answer was: D
Engage stakeholders to define the problem, affected persona, and outcome
- D2-002 · Domain II · II.8 — Define success criteria documentation-required
Trap: Engage stakeholders on the measurement method during Phase V evaluation
Why wrong: Wrong-answer trap. Working the measurement method with stakeholders is exactly what V.4 solution metrics work looks like, which makes the phrasing sound correct. Phase V instruments a criterion that already exists; it cannot be where the criterion first becomes measurable.
Correct answer was: D
Facilitate definition of a CSAT target with baseline, method, and time horizon
- D2-003 · Domain II · II.5 — Determine ROI documentation-required
Trap: Coordinate with governance to carry Trustworthy AI costs in their own budget
Why wrong: Wrong-answer trap. Charging Trustworthy AI work to a governance cost center is correct accounting when it is genuinely an enterprise control the project merely consumes, and cross-charging reads as clean. A project-specific audit, bias-monitoring, and review load is project cost, and hiding it distorts the number the sponsor decides on.
Correct answer was: C
Coordinate with finance and governance to fold Trustworthy AI costs into ROI
- D2-004 · Domain II · II.7 — Draft AI solution pattern-recognition
Trap: Recognition — identifying and understanding unstructured real-world data
Why wrong: Wrong-answer trap. Recognition identifies and understands unstructured real-world data, so it is the right pattern for reading a scanned claim form or verifying a face at onboarding. Real-time transaction records are structured, and the target is a behavioral deviation rather than an object to recognize.
Correct answer was: A
Patterns & Anomalies — detecting outliers and deviations from normal
- D2-005 · Domain II · II.3 — Conduct risk assessment cross-domain-link
Trap: Assess security risk now and revisit ethics at the operationalization gate
Why wrong: Wrong-answer trap. Security-first sequencing is correct when a breach is the only exposure and the ethics profile is already settled, and regulated data makes security the reflexive answer. Health data brings privacy, bias, and accountability in at the same moment, and the operationalization gate is far too late to meet them.
Correct answer was: C
Assess privacy, bias, regulatory, and accountability risk alongside technical risk
- D2-006 · Domain II · II.2 — Evaluate initial AI feasibility documentation-required
Trap: Practical deployment possible
Why wrong: Wrong-answer trap. Practical deployment possible is a Light 3, which makes the position look right, but it sits in the TECHNOLOGY row rather than the DATA row.
Correct answer was: C
Sufficient data quality
- D2-007 · Domain II · II.6 — Manage adoption/integration risks iteration-trigger
Trap: Escalate the concern to the sponsor as a risk requiring additional funding
Why wrong: Wrong-answer trap. Escalating for funding is the correct move once a mitigation plan exists and its cost exceeds what the PM controls, and escalation reads as proper governance. Nothing has been assessed or planned yet, so there is no scoped ask for the sponsor to fund.
Correct answer was: B
Log it as an adoption risk and build a mitigation plan with training and disclosure
- D2-008 · Domain II · II.4 — Develop AI project scope statement documentation-required
Trap: The delivery timeline with phase milestones and the target production date
Why wrong: Wrong-answer trap. Timeline and milestones dominate the scope conversation and do appear in the scope statement, which is what makes this tempting. Dates without in-scope and out-of-scope boundaries are precisely what invites scope creep.
Correct answer was: A
Problem, pattern, success criteria, boundaries, resources, risks, constraints
- D2-009 · Domain II · II.10 — Identify project resources cross-domain-link
Trap: IV.1 model technique selection, where the technique drives hardware need
Why wrong: Wrong-answer trap. Technique selection is genuinely shaped by available compute, which makes the causal story sound right, but IV.1 chooses the approach while IV.3 model training is where the resource constraint actually binds.
Correct answer was: D
III.4 workspace and infrastructure, IV.3 training compute, V.1 deployment plan
- D2-010 · Domain II · II.8 — Define success criteria go-no-go-gate
Trap: Coordinate with the data scientist to tune the model toward the documented target
Why wrong: Wrong-answer trap. Tuning toward a documented target is the correct IV.3 response when the target is realistic and the model underperforms it, and it reads as the constructive technical answer. It treats an optimistic business number as fixed and pressures the team to overfit toward it.
Correct answer was: C
Engage stakeholders to revise the criteria to what is achievable, then document it
- D2-011 · Domain II · II.1 — Identify problem to be solved documentation-required
Trap: Persona sits in adoption planning, where the user community is engaged
Why wrong: Wrong-answer trap. II.6 adoption planning engages the affected user community and is the most persona-driven task in the domain, which makes it a convincing home. Adoption planning targets a persona that problem identification already established.
Correct answer was: B
Persona sits in problem identification — who is affected shapes the problem
- D2-012 · Domain II · II.2 — Evaluate initial AI feasibility pattern-recognition
Trap: Proceed with caution and active risk management of the red item
Why wrong: Wrong-answer trap. Proceed with caution and active risk management is a genuine decision rule quoted exactly — it is the rule for multiple yellow lights. Applying it to a red is precisely the softening the framework exists to prevent.
Correct answer was: B
Stop and hold the project until the red item has been resolved
- D2-013 · Domain II · II.7 — Draft AI solution iteration-trigger
Trap: Coordinate with the data scientist to reselect the pattern, then brief stakeholders
Why wrong: Wrong-answer trap. Coordinating the reselection with the data scientist and briefing stakeholders afterward reads as collaborative and efficient, and it is the right sequence for a technique choice inside IV.1. A pattern change is a stakeholder decision taken before the fact, and this skips the II.4 scope impact assessment.
Correct answer was: C
Convene a stakeholder review of the pattern selection and assess the scope impact
- D2-014 · Domain II · II.9 — Support business case creation documentation-required
Trap: Sign off on the business case as the accountable owner leadership escalates to
Why wrong: Wrong-answer trap. Naming a single accountable owner is sound governance, and the PM does sign the artifacts they own such as the II.4 scope statement. II.9 is worded as support: signing off substitutes authority for the input gathering that is the actual task.
Correct answer was: C
Coordinate the business case, gathering finance, data science, and business input
- D2-015 · Domain II · II.8 — Define success criteria cross-domain-link
Trap: The V.4 production metrics and the V.7 contingency plan for a fairness incident
Why wrong: Wrong-answer trap. Production monitoring and a contingency plan are where fairness failures become visible and get handled, which makes this feel like the decisive control. Detecting a fairness violation on live traffic is the most expensive place to learn about it.
Correct answer was: B
Every stage: data requirements, evaluation, QA/QC, gate, monitoring, contingency
- D2-016 · Domain II · II.5 — Determine ROI documentation-required
Trap: Use the finance template and flag the AI items to stakeholders as assumptions
Why wrong: Wrong-answer trap. Flagging uncertain items as open assumptions is a legitimate estimating technique and the right move for figures that genuinely cannot be estimated yet, which is what makes it read as rigorous. Trustworthy AI costs are estimable here, and flagging them keeps material cost out of the number the sponsor decides on.
Correct answer was: C
Adapt the template to add AI value sources and the Trustworthy AI cost lines
- D2-017 · Domain II · II.3 — Conduct risk assessment documentation-required
Trap: Safety risk — harm arising from AI-driven decisions
Why wrong: Wrong-answer trap. Safety reads as a robotics or physical-systems concern, which makes it the easiest category to rule out on a software project. The ECO names it explicitly, and an AI decision that wrongly denies care or credit does real harm.
Correct answer was: B
Marketing-campaign risk — channel mix and advertising spend
- D2-018 · Domain II · II.10 — Identify project resources documentation-required
Trap: Hardware, licenses, and workspace
Why wrong: Wrong-answer trap. Hardware belongs on the list and licenses and workspace feel like its natural companions, which makes this the closest miss. The ECO pairs hardware with people and contractors, and the workspace itself is coordinated later at III.4 out of these resources.
Correct answer was: C
People, hardware, and contractors
- D2-019 · Domain II · II.6 — Manage adoption/integration risks cross-domain-link
Trap: Escalate the opt-out requirement to legal for a regulatory determination first
Why wrong: Wrong-answer trap. Escalating to legal is the right first step under I.4 when the regulatory basis for an obligation is genuinely unsettled, and it reads as careful governance. Consent and disclosure are Trustworthy AI obligations the project plans for regardless of what any single regulation compels.
Correct answer was: B
Document a consent mechanism with feasible alternatives in the adoption plan
- D2-020 · Domain II · II.7 — Draft AI solution pattern-recognition
Trap: Predictive Analytics & Decision Support — forecasting to guide decisions
Why wrong: Wrong-answer trap. Predictive Analytics & Decision Support forecasts to guide a human decision, and a bot suggesting a next best action makes it sound close. The chatbot's core job is the natural language exchange, not the forecast behind a suggestion.
Correct answer was: A
Conversational & Human Interaction — natural language dialogue with users
- D2-021 · Domain II · II.8 — Define success criteria documentation-required
Trap: Ask stakeholders to restate every success criterion as a quantitative metric
Why wrong: Wrong-answer trap. Demanding numbers is the correct correction when a stakeholder offers a criterion with no rating method and no threshold, and it sounds like the rigorous position. PMI treats structured qualitative criteria such as user trust and explainability as legitimate when they carry a defined rating method.
Correct answer was: B
Define measurable, time-bound criteria tied to the business objective now
- D2-022 · Domain II · II.4 — Develop AI project scope statement iteration-trigger
Trap: Coordinate with the data scientist to simplify the model to fit the timeline
Why wrong: Wrong-answer trap. Simplifying the model is real work a data scientist can do and the right IV.1 response when the technique is heavier than the problem needs. Scope is a stakeholder-owned commitment that a single technical change cannot renegotiate.
Correct answer was: B
Engage stakeholders on the options: rescope, extend, resource, or phase delivery
- D2-023 · Domain II · II.5 — Determine ROI iteration-trigger
Trap: Draw the added cost from contingency and report it in the phase status report
Why wrong: Wrong-answer trap. Drawing on contingency and reporting the draw is correct budget practice for a variance the ROI already anticipated, and the status report keeps it visible, which is what makes it look clean. A threefold increase is a change to the ROI itself, not a draw against it.
Correct answer was: A
Update the ROI with the actual data preparation cost and revisit the business case
- D2-024 · Domain II · II.1 — Identify problem to be solved iteration-trigger
Trap: Coordinate a hybrid build that keeps AI where the rules become complex
Why wrong: Wrong-answer trap. Hybrids are explicitly endorsed in CPMAI — separating the cognitive parts from the noncognitive parts is standard practice — which makes this the most methodologically literate option on the page. Building one now assumes an AI component is needed rather than testing whether it is.
Correct answer was: D
Loop back to II.1 with stakeholders to reassess whether the problem needs AI
- D2-025 · Domain II · II.7 — Draft AI solution documentation-required
Trap: The selected algorithm and the model family it belongs to, with justification
Why wrong: Wrong-answer trap. An algorithm with its model family and a justification is exactly what a IV.1 technique review should contain, and naming it sounds like the substantive technical answer. At II.7 it disconnects the proposed solution from the business problem the sponsor is being asked to evaluate.
Correct answer was: D
The AI pattern, persona, key data needs, technique direction, criteria, and risks
- D2-026 · Domain II · II.3 — Conduct risk assessment cross-domain-link
Trap: Data protection, access control, and breach exposure across the model pipeline
Why wrong: Wrong-answer trap. Data protection, access control, and breach exposure is the right enumeration for the security risk category the ECO names in the very same task statement, which makes it read as an in-scope answer. It is the sibling category, not what ethics risk means.
Correct answer was: C
Bias and fairness, human accountability, workforce impact, consent, disclosure
- D2-027 · Domain II · II.10 — Identify project resources iteration-trigger
Trap: Coordinate with operations, which owns cloud capacity across the organization
Why wrong: Wrong-answer trap. Operations does own cloud capacity and becomes the right partner once the solution is running under V.3 model governance, which makes the handoff sound like correct routing. Pre-deployment project resources are the PM's II.10 responsibility to resolve.
Correct answer was: D
Work the alternatives with the platform team and take any impact to stakeholders
- D2-028 · Domain II · II.6 — Manage adoption/integration risks documentation-required
Trap: The product owner owns adoption and engages users through the product roadmap
Why wrong: Wrong-answer trap. The product owner is closest to the users and holds the roadmap, which makes this the most credible single-owner answer and the right one for feature adoption on a product backlog. Adoption risk on an AI project is a project risk the ECO assigns to the PM.
Correct answer was: C
The PM coordinates adoption risk across the functions executing mitigation
- D2-029 · Domain II · II.8 — Define success criteria go-no-go-gate
Trap: Record GO and document a follow-up action to revisit the criteria in Phase IV
Why wrong: Wrong-answer trap. Passing a gate while tracking a concern as a documented follow-up is orderly, common project practice and the right handling for an issue outside the gate's scope. It lets the project run on criteria already known to be wrong, and every downstream evaluation reads from them.
Correct answer was: A
Iterate back to II.8 to revise the criteria with stakeholders, then rerun the gate
- D2-030 · Domain II · II.5 — Determine ROI documentation-required
Trap: When stakeholders formally request a revision to the approved business case
Why wrong: Wrong-answer trap. Waiting for a formal stakeholder request respects change-control governance and is the right posture for a baselined scope statement. Stakeholders cannot know an ROI update is needed; surfacing material change is the PM's II.5 responsibility.
Correct answer was: D
Continuously, whenever cost, scope, or benefit assumptions materially change
- D3-001 · Domain III · III.1 — Define required data oversight-vs-doing
Trap: Have the data engineer document access terms and cost for each candidate source
Why wrong: Wrong-answer trap — documenting access terms and cost is genuine III.3 source work and looks like progress, but sourcing follows the requirements, and III.1 has not yet produced anything to source against.
Correct answer was: D
Produce a documented data requirements specification tied to Phase I criteria
- D3-002 · Domain III · III.8 — Determine if data meets solution needs go-no-go-gate
Trap: Recommend a NO-GO and iterate back to Phase I to revisit the business problem
Why wrong: Wrong-answer trap — a NO-GO looping back to Phase I is the right call when the business problem itself is wrong, which makes it sound decisive, but the PM does not pick the outcome before the gate convenes and a sourcing gap loops back inside Phase II.
Correct answer was: D
Convene a documented gate where stakeholders make the GO or NO-GO decision
- D3-003 · Domain III · III.6 — Check data privacy, compliance, and access cross-domain-link
Trap: Record the agreement as a planning assumption and continue Phase II as scheduled
Why wrong: Wrong-answer trap — recording a dependency as a planning assumption is standard practice for items off the critical path, but this one gates data access itself, and negotiated agreements can fail or slip well past six weeks.
Correct answer was: C
Document the six-week dependency as a risk and put the options to stakeholders
- D3-004 · Domain III · III.5 — Gather required data iteration-trigger
Trap: Coordinate with the platform team to restore the decommissioned source system
Why wrong: Wrong-answer trap — asking the platform team to restore a system is a coordinated, governed-looking move and is right when a system was retired in error, but this one was decommissioned deliberately under another team's change control.
Correct answer was: D
Loop back to III.3 for alternatives and reassess the impact on the III.8 gate
- D3-005 · Domain III · III.2 — Identify data SMEs oversight-vs-doing
Trap: Coordinate with the data engineer to trace each field back to its source system
Why wrong: Wrong-answer trap — lineage tracing is disciplined engineering work and answers where a field came from, but the stem asks what the field means, which is a stewardship question rather than a pipeline one.
Correct answer was: A
Identify current data stewards or domain experts who can confirm field meanings
- D3-006 · Domain III · III.7 — Oversee data evaluation documentation-required
Trap: Have the data steward sign off on the fitness conclusion before the gate convenes
Why wrong: Wrong-answer trap — steward sign-off is real governance and belongs to the III.6 access and compliance checklist, but it does not stand in for the III.7 evaluation report the gate consumes.
Correct answer was: D
Confirm a documented data evaluation report exists before scheduling the gate
- D3-007 · Domain III · III.4 — Coordinate AI workspace and infrastructure escalation-trigger
Trap: Document the shortfall as a risk in the register and revisit it at the gate
Why wrong: Wrong-answer trap — logging it in the risk register and carrying it to the gate is textbook risk discipline, but the shortfall blocks Phase II work now and the gate is not the forum that allocates capacity.
Correct answer was: D
Escalate the capacity gap to leadership with cost, scope, and schedule options
- D3-008 · Domain III · III.8 — Determine if data meets solution needs go-no-go-gate
Trap: Take the production-retraining plan to the gate as the documented GO rationale
Why wrong: Wrong-answer trap — continuous retraining is a real operations capability and packaging it as gate rationale looks rigorous, but it asks the gate to approve building on data that does not represent production.
Correct answer was: C
Convene the gate and recommend NO-GO, iterating back to re-source current data
- D3-009 · Domain III · III.9 — Convey data understanding to leadership documentation-required
Trap: Convene the data SMEs to walk through the evaluation findings before Phase III
Why wrong: Wrong-answer trap — walking the SMEs through the findings is genuine consultation and belongs inside III.7 evaluation, but that audience has already served its purpose and cannot authorize the next phase.
Correct answer was: B
Brief leadership on Phase II findings and the gate decision before Phase III
- D3-010 · Domain III · III.6 — Check data privacy, compliance, and access cross-domain-link
Trap: Have the data steward approve pseudonymizing the dataset before any transfer occurs
Why wrong: Wrong-answer trap — steward approval of pseudonymization wears both governance and technical clothing, but pseudonymized data remains personal data under GDPR and stewards do not make transfer determinations.
Correct answer was: D
Coordinate a GDPR cross-border transfer assessment with privacy and legal first
- D3-011 · Domain III · III.1 — Define required data phase-identification
Trap: Phase I — Business Understanding, where the business problem is first framed
Why wrong: Wrong-answer trap — Phase I supplies the success criteria and AI pattern that constrain the data, which makes it feel like the origin, but defining required data itself belongs to Phase II.
Correct answer was: C
Phase II — Data Understanding, where required data is defined for the use case
- D3-012 · Domain III · III.3 — Identify data sources and locations documentation-required
Trap: Coordinate with procurement and hold the inventory until renewal is decided
Why wrong: Wrong-answer trap — coordinating with procurement and waiting for certainty sounds careful, but the renewal sits six months out and III.5 gathers against the inventory long before then.
Correct answer was: D
Document the source and escalate the contract expiry as a project dependency
- D3-013 · Domain III · III.7 — Oversee data evaluation cross-domain-link
Trap: Coordinate the deployment metrics so fairness is monitored once in production
Why wrong: Wrong-answer trap — wiring fairness into the V.4 solution metrics is real governance and will be needed eventually, but production monitoring makes affected users the detection mechanism for a skew already visible now.
Correct answer was: A
Document the bias finding and take the mitigation options to stakeholders
- D3-014 · Domain III · III.8 — Determine if data meets solution needs pattern-recognition
Trap: GO to Phase III, NO-GO to cancel, or On-Hold pending more data gathering
Why wrong: Wrong-answer trap — GO and NO-GO is the right pair, which makes this look almost correct, but On-Hold is not a defined outcome and a NO-GO sends the project back to an earlier phase rather than cancelling it.
Correct answer was: A
GO to Phase III, or NO-GO — pause and iterate back to an earlier phase
- D3-015 · Domain III · III.4 — Coordinate AI workspace and infrastructure oversight-vs-doing
Trap: Submit the request to the security review board and provision once it clears
Why wrong: Wrong-answer trap — routing the request through the security review board is genuine governance and does cover the new cloud region, but it leaves the capital cost unapproved by anyone holding spending authority.
Correct answer was: C
Coordinate with platform, security, and finance before capacity is provisioned
- D3-016 · Domain III · III.6 — Check data privacy, compliance, and access phase-identification
Trap: Approve Phase III timing, since that is when the data is actively transformed
Why wrong: Wrong-answer trap — Phase III timing is right for the handling controls applied during transformation, such as masking and access scoping, but the eligibility question has to be settled before any preparation effort is spent.
Correct answer was: B
Move the checks into Phase II so disqualifying issues surface before prep
- D3-017 · Domain III · III.5 — Gather required data documentation-required
Trap: Accept the completion report and record the handoff in the project decision log
Why wrong: Wrong-answer trap — logging the handoff in the decision log is documentation discipline and feels like the governed choice, but a completion report is a claim and the inventory and staged data are the evidence behind it.
Correct answer was: B
Confirm the source inventory status and that staged data matches requirements
- D3-018 · Domain III · III.1 — Define required data iteration-trigger
Trap: Record the pattern change in the decision log on the data scientist's advice
Why wrong: Wrong-answer trap — logging the decision makes the change traceable and the data scientist is genuinely closest to technical fit, but pattern selection carries scope and business-case consequences that stakeholders own.
Correct answer was: D
Loop back to Phase I so stakeholders reconsider the AI pattern selection
- D3-019 · Domain III · III.8 — Determine if data meets solution needs go-no-go-gate
Trap: Record a conditional GO with privacy sign-off tracked as an open gate action
Why wrong: Wrong-answer trap — a conditional GO with a tracked action is how many organizations keep schedule while governance closes, but it starts Phase III on data whose lawful use is unconfirmed and leaves the condition owned by nobody.
Correct answer was: B
NO-GO — hold the phase transition until the privacy plan has legal sign-off
- D3-020 · Domain III · III.7 — Oversee data evaluation oversight-vs-doing
Trap: Profiling tooling evaluates the data and the PM signs off the generated report
Why wrong: Wrong-answer trap — profiling tools do produce much of the evidence, which makes this sound current, but tooling cannot judge whether a field means what the business believes it means.
Correct answer was: A
The data scientist, SMEs, and stewards evaluate; the PM oversees and documents
- D3-021 · Domain III · III.6 — Check data privacy, compliance, and access escalation-trigger
Trap: Have the data engineer pseudonymize the dataset under the steward's direction
Why wrong: Wrong-answer trap — pseudonymization under steward direction wears both technical and governance clothing, but it is reversible, still yields personal data, and leaves the consent defect untouched.
Correct answer was: B
Treat it as a compliance incident: stop use and escalate to legal and compliance
- D3-022 · Domain III · III.2 — Identify data SMEs pattern-recognition
Trap: Steward and custodian are interchangeable titles for the same accountability
Why wrong: Wrong-answer trap — the titles are used loosely in many organizations, which makes conflating them feel harmless, but PMI tests them as distinct accountabilities.
Correct answer was: D
Stewards enforce policy and curate data; custodians handle storage and transfer
- D3-023 · Domain III · III.5 — Gather required data iteration-trigger
Trap: Approve the synthetic data plan with the data scientist defining the quality checks
Why wrong: Wrong-answer trap — synthetic augmentation with defined quality checks is a legitimate option and looks properly governed, but it is one of several alternatives carrying different costs for stakeholders to weigh.
Correct answer was: C
Pause and put synthetic data, more sourcing, or a narrower scope to stakeholders
- D3-024 · Domain III · III.4 — Coordinate AI workspace and infrastructure cross-domain-link
Trap: Record the manual install steps in the environment configuration baseline
Why wrong: Wrong-answer trap — capturing the steps in the configuration baseline is real engineering discipline and would be right for a throwaway spike environment, but a documented manual process is still manual and still breaks reproducibility.
Correct answer was: B
Coordinate a supported custom image with platform, or revisit the technique
- D3-025 · Domain III · III.9 — Convey data understanding to leadership documentation-required
Trap: The full data evaluation report the data scientist produced during Phase II
Why wrong: Wrong-answer trap — the evaluation report is the most substantive artifact available, which makes forwarding it feel thorough, but it is III.7's output written for a technical audience.
Correct answer was: A
Data state, gate decision, risks, recommendations, and scope or schedule impact
- D3-026 · Domain III · III.1 — Define required data documentation-required
Trap: Record the missing linkage as a documented gap and move on to source work
Why wrong: Wrong-answer trap — recording a documented gap is proper practice for something that cannot be fixed yet, but this linkage is fixable now at almost no cost, and III.3 sourcing depends on it.
Correct answer was: B
Revise the specification to tie each field to the Phase I pattern and criteria
- D3-027 · Domain III · III.3 — Identify data sources and locations oversight-vs-doing
Trap: Record the source in the inventory once the data scientist has started evaluating it
Why wrong: Wrong-answer trap — documenting after the fact keeps the team moving and eventually produces the record, but the inventory exists to gate use rather than to narrate it.
Correct answer was: C
Verify license, provenance, bias, and privacy fit before it enters the inventory
- D3-028 · Domain III · III.7 — Oversee data evaluation documentation-required
Trap: Have the data scientist define and sign off the evaluation criteria
Why wrong: Wrong-answer trap — letting the evaluator set and sign off the bar sounds like empowering the expert, and is right for tooling choices, but it lets one person define and then pass their own test.
Correct answer was: B
Provide a documented evaluation rubric tied to the project's success criteria
- D3-029 · Domain III · III.8 — Determine if data meets solution needs go-no-go-gate
Trap: GO with the volume shortfall recorded as a risk for Phase III to manage
Why wrong: Wrong-answer trap — recording a known shortfall as a managed Phase III risk looks like disciplined risk practice, but it converts a gate finding into debt and starts a phase the data cannot yet support.
Correct answer was: B
NO-GO — take pretrained models, augmentation, or sourcing to stakeholders
- D3-030 · Domain III · III.6 — Check data privacy, compliance, and access cross-domain-link
Trap: Escalate to the sponsor to have the training requirement waived for the team
Why wrong: Wrong-answer trap — escalating to the sponsor is the right move when a requirement is disproportionate to the project, but sponsors fund and prioritize; they do not waive the access conditions a data steward sets.
Correct answer was: A
Build the steward's required training into the schedule rather than skipping it
- D3-031 · Domain III · III.4 — Coordinate AI workspace and infrastructure documentation-required
Trap: Track compute, storage, and network spend against the project's approved budget
Why wrong: Wrong-answer trap — cost tracking is genuinely part of coordinating infrastructure, but treating spend against budget as the task reduces III.4 to its smallest component.
Correct answer was: B
Coordinate with platform, security, and engineering so the workspace is ready
- D3-032 · Domain III · III.5 — Gather required data escalation-trigger
Trap: Have the data engineer redact identifying fields under the steward's direction
Why wrong: Wrong-answer trap — redaction under steward direction addresses privacy risk and looks properly governed, but the contract restricts the purpose of use, which redaction does nothing to change.
Correct answer was: C
Escalate to legal and contract management and remove the data pending review
- D3-033 · Domain III · III.2 — Identify data SMEs oversight-vs-doing
Trap: Coordinate with the SME's manager to route questions through a team inbox
Why wrong: Wrong-answer trap — routing questions through the SME's team inbox is a coordinated arrangement and is right when the named SME is genuinely unavailable, but it produces no commitment, no cadence, and no record of what was confirmed.
Correct answer was: C
Set up a documented SME engagement plan with cadence and async methods
- D3-034 · Domain III · III.1 — Define required data pattern-recognition
Trap: Determine how the required data will be aggregated across its sources
Why wrong: Wrong-answer trap — aggregation sounds like transformation work belonging to Phase III, which makes it look out of place here, but determining the means of aggregating the data is one of the four activities.
Correct answer was: D
Build the ingestion pipelines that load the identified data for modeling
- D3-035 · Domain III · III.8 — Determine if data meets solution needs go-no-go-gate
Trap: Record GO with the quality and regulatory concerns tracked as open actions
Why wrong: Wrong-answer trap — logging concerns as tracked gate actions is how many organizations protect a schedule, but it records agreement that does not exist and starts Phase III with the quality and regulatory questions still open.
Correct answer was: B
Facilitate a criteria-based review of each concern and take the trade-offs up
- D3-036 · Domain III · III.6 — Check data privacy, compliance, and access documentation-required
Trap: Have the data engineer anonymize the dataset to remove it from disclosure scope
Why wrong: Wrong-answer trap — anonymization can reduce privacy obligations in some cases, which makes it a tempting shortcut, but it does not remove a disclosure requirement legal has already identified.
Correct answer was: D
Hold preparation until the disclosure is in place and surface the schedule hit
- D3-037 · Domain III · III.5 — Gather required data iteration-trigger
Trap: Approve imputation with the data scientist documenting the method used
Why wrong: Wrong-answer trap — imputation with a documented method is standard practice and often the eventual answer, but approving it before evaluation assumes it preserves the quality and fairness the project committed to.
Correct answer was: A
Send the missingness to III.7 evaluation and take the options to stakeholders
- D3-038 · Domain III · III.3 — Identify data sources and locations cross-domain-link
Trap: Coordinate with the data engineer to strip identifiers from prompts first
Why wrong: Wrong-answer trap — coordinating identifier stripping addresses the obvious risk and looks like a controlled design, but free-text prompts routinely carry personal data that field-level redaction never sees.
Correct answer was: D
Assess the privacy, compliance, and disclosure implications with legal first
- D3-039 · Domain III · III.4 — Coordinate AI workspace and infrastructure escalation-trigger
Trap: Rework the workspace around the storage, compute, and services already approved
Why wrong: Wrong-answer trap — reworking the design around approved services is the low-friction path and is right once an exception is denied, but adopting it before the request is even made may compromise the solution for no reason.
Correct answer was: A
Coordinate the exception request with security, architecture, and compliance
- D3-040 · Domain III · III.8 — Determine if data meets solution needs iteration-trigger
Trap: Recommend NO-GO, iterate back to III.1, and respecify the required data
Why wrong: Wrong-answer trap — a NO-GO looping back to the data requirements is the standard response to most gate problems, which is what makes it tempting, but the data already meets those requirements and it is the II.8 success criteria that are in doubt.
Correct answer was: B
Recommend NO-GO, iterate back to II.8 to revise the criteria, then rerun
- D3-041 · Domain III · III.1 — Define required data cross-domain-link
Trap: Record fairness as a risk and address it when the sources are chosen
Why wrong: Wrong-answer trap — tracking fairness in the risk register looks organized and is right once representation is already specified, but a requirement sitting outside the specification does not shape what actually gets sourced.
Correct answer was: B
Specify demographic representation and bias-measurement needs in the spec
- D3-042 · Domain III · III.5 — Gather required data documentation-required
Trap: A privacy impact assessment covering each of the gathered sources
Why wrong: Wrong-answer trap — privacy assessment is genuinely required in Phase II, which makes it feel like a gathering output, but it is III.6's deliverable rather than III.5's.
Correct answer was: A
An updated source inventory with status, plus the staged dataset
- D3-043 · Domain III · III.7 — Oversee data evaluation pattern-recognition
Trap: Validity — whether values conform to defined rules and schema
Why wrong: Wrong-answer trap — validity sounds like a generic quality adjective rather than a named dimension, but it is one of the seven and tests conformance to defined rules.
Correct answer was: D
Recency — whether the newest records dominate the dataset
- D3-044 · Domain III · III.2 — Identify data SMEs oversight-vs-doing
Trap: Yes — the custodian is the legal owner of record for the data
Why wrong: Wrong-answer trap — legal-ownership language sounds authoritative and is used loosely in practice, but PMI defines the custodian as explicitly not the owner.
Correct answer was: B
No — the custodian administers the data but is not its owner
- D3-045 · Domain III · III.6 — Check data privacy, compliance, and access phase-identification
Trap: I.2 transparency over the data and algorithm selection choices
Why wrong: Wrong-answer trap — transparency does have a data dimension, which makes it plausible, but I.2 concerns disclosing how data and algorithms were selected rather than whether the data may lawfully be used.
Correct answer was: A
I.1 privacy and security plan, with I.4 regulatory compliance
- D3-046 · Domain III · III.8 — Determine if data meets solution needs go-no-go-gate
Trap: Document the hallway conversation and file it as the gate decision record
Why wrong: Wrong-answer trap — writing it up afterwards produces a document and feels like closing the gap, but it records a decision the review never formally made.
Correct answer was: A
Convene a formal gate review and produce the documented decision
- D3-047 · Domain III · III.1 — Define required data documentation-required
Trap: Validation, test, and inference data are specified in Phase IV, not Phase II
Why wrong: Wrong-answer trap — validation, test, and inference data are all used after Phase II, which makes deferring their specification sound like correct phase discipline, but requirements for every category are set in Phase II.
Correct answer was: C
Training, validation, test, and production-inference data all get specified
- D3-048 · Domain III · III.3 — Identify data sources and locations iteration-trigger
Trap: Have the data engineer estimate the shortfall's impact on model accuracy
Why wrong: Wrong-answer trap — an impact estimate sounds like the analytical first step, but the engineer can size the shortfall without addressing it, and the decision needs stakeholder context the estimate does not carry.
Correct answer was: C
Loop back to III.3 to evaluate additional internal sources and revise the inventory
- D3-049 · Domain III · III.7 — Oversee data evaluation cross-domain-link
Trap: Escalate to stakeholders that expected drift may invalidate the use case
Why wrong: Wrong-answer trap — escalating drift to stakeholders sounds appropriately transparent, but evaluation found the data fit for use, and expected drift is design input for operations rather than a threat to the use case.
Correct answer was: A
Document the drift expectations now so they shape the Phase VI design work
- D3-050 · Domain III · III.4 — Coordinate AI workspace and infrastructure documentation-required
Trap: Cost estimates for compute and storage, signed off by the finance partner
Why wrong: Wrong-answer trap — cost is the attribute stakeholders ask about most and a signed estimate looks like a controlled artifact, but an estimate without provisioning status tracks nothing.
Correct answer was: B
A provisioning plan and status tracker with security and access sign-off
- D3-051 · Domain III · III.5 — Gather required data oversight-vs-doing
Trap: Ingestion tooling gathers the data and the PM documents the results
Why wrong: Wrong-answer trap — ingestion is heavily automated in practice, which makes this sound current, but the tooling runs under engineering ownership and documenting output is not the same as coordinating the work.
Correct answer was: A
Data engineering gathers against the inventory while the PM coordinates
- D3-052 · Domain III · III.9 — Convey data understanding to leadership escalation-trigger
Trap: Have the data scientist present the quality, volume, and drift findings
Why wrong: Wrong-answer trap — having the data scientist present the technical findings looks appropriately humble and they do own the detail, but schedule and delivery commitments are the PM's accountability.
Correct answer was: C
State the data risks honestly and offer scope, resource, or date options
- D3-053 · Domain III · III.6 — Check data privacy, compliance, and access pattern-recognition
Trap: Pseudonymization is the stronger control because the mapping is protected
Why wrong: Wrong-answer trap — pseudonymization feels stronger because it sounds more technical and preserves a controlled key, but reversibility is exactly what makes it the weaker protection.
Correct answer was: B
Anonymization is irreversible; pseudonymized data can be re-identified
- D3-054 · Domain III · III.1 — Define required data iteration-trigger
Trap: Coordinate platform, engineering, and security to build the real-time collection path
Why wrong: Wrong-answer trap — coordinating the teams who would build the real-time path looks like exactly the III.4 coordination the PM owns, but it commits cost and scope that no stakeholder has approved.
Correct answer was: A
Take the capability gap to stakeholders to invest, revise the criteria, or cut scope
- D3-055 · Domain III · III.3 — Identify data sources and locations documentation-required
Trap: Have the data engineer populate the column with the expected access mechanism
Why wrong: Wrong-answer trap — having the engineer fill the column from the access paths they expect completes the document and looks like ordinary engineering judgment, but a presumed access route can bypass the controls the real one enforces.
Correct answer was: C
Resolve the access mechanism for every entry before the inventory is final
- D3-056 · Domain III · III.7 — Oversee data evaluation cross-domain-link
Trap: Trustworthy-AI verification belongs to the IV.6 gate before operationalization
Why wrong: Wrong-answer trap — trustworthy-AI concerns are certainly verified again at IV.6 before operationalization, which makes that gate sound like their home, but discovering unusable data then wastes every phase in between.
Correct answer was: B
Bias, privacy, and transparency checks belong inside the evaluation rubric
- D3-057 · Domain III · III.2 — Identify data SMEs documentation-required
Trap: Contact details, working hours, and time zone for each identified SME
Why wrong: Wrong-answer trap — contact details, hours, and time zone are the practical things a PM reaches for first, but knowing how to reach someone says nothing about what they have agreed to support.
Correct answer was: B
Data element, SME role and name, engagement scope, cadence, and backup
- D3-058 · Domain III · III.8 — Determine if data meets solution needs go-no-go-gate
Trap: Proceed as a pilot and record the scope mismatch in the project's decision log
Why wrong: Wrong-answer trap — recording the mismatch honestly and continuing looks like transparency, but the record does not change what the pilot's results will be worth to the production decision.
Correct answer was: A
Loop back to Phase I and redefine the effort explicitly as a proof of concept
- D3-059 · Domain III · III.6 — Check data privacy, compliance, and access iteration-trigger
Trap: Convene stakeholders to drop the restricted element from the requirements
Why wrong: Wrong-answer trap — taking the removal to stakeholders is properly governed and is right once the policy owners refuse an extension, but dropping a required element before asking forfeits data the project may be entitled to use.
Correct answer was: A
Escalate to the data steward and policy owners, or revisit the requirement
- D3-060 · Domain III · III.4 — Coordinate AI workspace and infrastructure phase-identification
Trap: Infrastructure decisions are not part of identifying data needs at all
Why wrong: Wrong-answer trap — infrastructure feels like an engineering topic rather than a data-needs one, but III.4 names workspace and infrastructure coordination explicitly.
Correct answer was: D
Workspace, environments, and pipelines are III.4; production sits with V.1
- D3-061 · Domain III · III.1 — Define required data documentation-required
Trap: Coverage of the pattern, attributes, sources, and the aggregation plan
Why wrong: Wrong-answer trap — full content coverage is most of what the specification needs and sounds like a complete answer, but III.1 closes on agreement as well as content.
Correct answer was: D
Full content coverage plus sign-off from SMEs, data science, and stakeholders
- D3-062 · Domain III · III.5 — Gather required data cross-domain-link
Trap: Have the data engineer flag and exclude records without AI-use consent
Why wrong: Wrong-answer trap — excluding non-consenting records is very likely part of the eventual answer, which makes it tempting, but an engineer applying consent logic without legal review is deciding a regulatory question.
Correct answer was: C
Assess the consent tiers with privacy and legal before gathering proceeds
- D3-063 · Domain III · III.8 — Determine if data meets solution needs documentation-required
Trap: The decision, the date, and the attendees recorded in the project log
Why wrong: Wrong-answer trap — decision, date, and attendees is what a project log normally carries and looks like a complete record, but without findings and rationale the decision cannot be audited or revisited.
Correct answer was: A
A gate decision document with findings, rationale, sign-off, and next steps
- D3-064 · Domain III · III.4 — Coordinate AI workspace and infrastructure oversight-vs-doing
Trap: Approve the install, since evaluation is internal and not a commercial product
Why wrong: Wrong-answer trap — internal-use-only reasoning is a widespread assumption, but commercial-use clauses usually turn on the organization being commercial rather than on the output being sold.
Correct answer was: C
Take the license terms to legal and procurement and weigh the alternatives
- D3-065 · Domain III · III.3 — Identify data sources and locations pattern-recognition
Trap: Whether the source is internal to the organization or external
Why wrong: Wrong-answer trap — source type reads as background detail rather than a decision input, but internal versus external changes the access, cost, and legal picture entirely.
Correct answer was: C
The development methodology the source-owning team works in
- D3-066 · Domain III · III.6 — Check data privacy, compliance, and access documentation-required
Trap: Privacy, legal, and the data scientist who will actually use the data
Why wrong: Wrong-answer trap — privacy and legal are the right authorities and the data scientist knows the intended use, which makes this set look complete, but the data scientist signs nothing and compliance and the steward are missing.
Correct answer was: D
Privacy, legal, compliance, the data steward, and the project sponsor
- D3-067 · Domain III · III.2 — Identify data SMEs iteration-trigger
Trap: Record the gap in the risk register and have assumptions documented
Why wrong: Wrong-answer trap — a risk-register entry with documented assumptions is proper practice once stakeholders have accepted the knowledge gap, but recording it first accepts that risk on their behalf.
Correct answer was: D
Raise the SME gap as a project risk and take the options to stakeholders
- D3-068 · Domain III · III.9 — Convey data understanding to leadership pattern-recognition
Trap: The key data risks identified over the course of Phase II evaluation
Why wrong: Wrong-answer trap — risks feel like internal project-management detail, but leadership visibility into data risk is the main reason III.9 exists.
Correct answer was: B
Field-level schemas and the queries used to profile the source tables
- D3-069 · Domain III · III.4 — Coordinate AI workspace and infrastructure iteration-trigger
Trap: Escalate to procurement to expedite the other department's contract renewal
Why wrong: Wrong-answer trap — pressing procurement to expedite the renewal targets the actual cause and looks like decisive coordination, but the contract belongs to another department and the project still needs a decision either way.
Correct answer was: C
Surface the delay as a risk with alternatives and let stakeholders decide
- D3-070 · Domain III · III.1 — Define required data oversight-vs-doing
Trap: Have the data scientist document the identified data and circulate it
Why wrong: Wrong-answer trap — circulating a documented list looks like the specification step, but it skips the traceability to success criteria and the multi-party sign-off that closes III.1.
Correct answer was: D
Require a signed specification linking criteria to attributes and sources
- D3-071 · Domain III · III.5 — Gather required data documentation-required
Trap: On a weekly cycle, matching the project's regular status reporting rhythm
Why wrong: Wrong-answer trap — a weekly rhythm matches normal status reporting and sounds disciplined, but during active gathering it can leave a blocked critical source unaddressed for days.
Correct answer was: A
Continuously as gathering proceeds, so blockers surface while they matter
- D3-072 · Domain III · III.7 — Oversee data evaluation documentation-required
Trap: Issue the report and add the bias section once the tooling becomes available
Why wrong: Wrong-answer trap — appending the section later sounds like a reasonable sequencing compromise, but the gate consumes the report before the appendix would ever arrive.
Correct answer was: A
Keep bias measurement in scope and close the tooling gap before sign-off
- D3-073 · Domain III · III.6 — Check data privacy, compliance, and access go-no-go-gate
Trap: GO conditionally, with the legal assessment tracked as an open action
Why wrong: Wrong-answer trap — a conditional GO protects the schedule and looks like a managed compromise, but it starts Phase III work whose legal basis is unresolved and leaves the condition owned by nobody.
Correct answer was: D
NO-GO — hold until legal concludes or leadership accepts the risk
- D3-074 · Domain III · III.2 — Identify data SMEs cross-domain-link
Trap: Have the data scientist represent the business perspective in SME sessions
Why wrong: Wrong-answer trap — the data scientist does sit between both worlds, which makes them look like an efficient single point of contact, but they consume the data rather than being an authority on its business meaning.
Correct answer was: D
Engage both, since structure and business meaning are validated separately
- D3-075 · Domain III · III.8 — Determine if data meets solution needs iteration-trigger
Trap: GO and add the new risks to the Phase III risk register for management
Why wrong: Wrong-answer trap — carrying the new risks into the Phase III risk register sounds like normal practice, but it commits the phase before anyone has judged whether the approach still holds.
Correct answer was: C
Convene a documented II.3 risk reassessment with stakeholders before the GO/NO-GO call
- D4-001 · Domain IV · IV.5 — Verify data quality (GATE) go-no-go-gate
Trap: Convene the III.8 gate and record the stakeholders' decision on data availability
Why wrong: Wrong-answer trap. III.8 closes Phase II by asking whether the right raw data exists at all, and it is the correct gate to convene at that boundary — but this team has already built and run the preparation pipelines, so that question is behind them.
Correct answer was: D
Convene the IV.5 gate and record the stakeholders' decision on the prepared data
- D4-002 · Domain IV · IV.3 — Manage AI/ML model training iteration-trigger
Trap: Pause the run and reopen the IV.1 technique decision with the stakeholders
Why wrong: Wrong-answer trap. Reopening IV.1 is a legitimate outcome and comes properly dressed as a stakeholder decision, which is what makes it tempting, but it commits the project to a technique explanation nobody has established.
Correct answer was: A
Pause the run and establish root cause before deciding how the project proceeds
- D4-003 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) go-no-go-gate
Trap: Convene the V.3 governance review and record the decision to operate the model
Why wrong: Wrong-answer trap. V.3 model governance is a real documented decision forum and sounds like the deployment authority, but it governs a model the organization already operates; it inherits the GO rather than issuing it.
Correct answer was: B
Convene the IV.6 gate and record the stakeholders' decision on model readiness
- D4-004 · Domain IV · IV.5 — Verify data quality (GATE) pattern-recognition
Trap: The same gate, since both close with a documented Go/No-Go decision and sign-off
Why wrong: Wrong-answer trap. Shared outcome vocabulary is the strongest reason candidates merge these two, but every CPMAI gate closes with a Go/No-Go decision and sign-off, so that structure says nothing about which artifact is being judged.
Correct answer was: C
Distinct gates: the earlier asks if the right data exists, the later if it can train
- D4-005 · Domain IV · IV.1 — Oversee AI/ML model technique(s) documentation-required
Trap: Document the post-hoc explainability plan and have the data scientist apply it
Why wrong: Wrong-answer trap. Post-hoc explainability is a legitimate approach and the right thing to document once stakeholders accept it as sufficient, but whether it clears the client's bar is their judgment and it has to come before the technique locks in.
Correct answer was: D
Document the explainability trade-off and have the client stakeholders decide
- D4-006 · Domain IV · IV.2 — Oversee AI/ML model QA/QC cross-domain-link
Trap: Document the bias finding and have the data scientist add a fairness layer
Why wrong: Wrong-answer trap. Post-processing fairness correction is a real mitigation and exactly what gets implemented once stakeholders select it, but a PM authorizing it directly turns a governance question into a technical patch.
Correct answer was: C
Document the bias finding and have stakeholders choose remediation before IV.6
- D4-007 · Domain IV · IV.3 — Manage AI/ML model training iteration-trigger
Trap: Pause and reopen the IV.1 technique choice with stakeholders before the next run
Why wrong: Wrong-answer trap. Changing algorithm family is a legitimate outcome of the review and arrives as a proper stakeholder decision, which is what makes it convincing, but choosing it first assumes the plateau is a technique problem rather than a data one.
Correct answer was: B
Pause and review the iteration trajectory before deciding whether to continue
- D4-008 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) go-no-go-gate
Trap: Record GO and have the ML engineer optimize the model for the CPU instances
Why wrong: Wrong-answer trap. CPU optimization is plausible and may succeed, and it is exactly the work a loop back to IV.1 would authorize — but doing it under a GO settles the question the gate exists to decide.
Correct answer was: A
Record NO-GO on operational fit; stakeholders pick the V.1 or IV.1 loop
- D4-009 · Domain IV · IV.4 — Manage data transformation to conduct data preparation oversight-vs-doing
Trap: Automated tooling performs the transformation; the PM reviews the prepared dataset
Why wrong: Wrong-answer trap. Tooling does much of the mechanical work, which makes this sound like modern practice, but tools run inside an engineered pipeline that people own, version, and document.
Correct answer was: B
Data engineering executes against the Data Preparation Plan; the PM coordinates it
- D4-010 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) go-no-go-gate
Trap: Confirm that deployment artifacts may be built now and released once GO lands
Why wrong: Wrong-answer trap. Holding the release sounds like it preserves the gate, but building deployment artifacts is V.2 execution that GO authorizes; drafting the V.1 deployment plan is the part that may legitimately proceed earlier.
Correct answer was: A
Confirm that operationalization starts only once the IV.6 gate returns GO
- D4-011 · Domain IV · IV.1 — Oversee AI/ML model technique(s) documentation-required
Trap: Yes — an automated, fully logged selection process satisfies the requirement
Why wrong: Wrong-answer trap. Automated tools do emit detailed run logs, and logs feel like documentation, but a run log is not a justification against the AI pattern, success criteria, and operational constraints.
Correct answer was: C
No — AutoML automates the technical choice; its output still feeds IV.1 records
- D4-012 · Domain IV · IV.5 — Verify data quality (GATE) go-no-go-gate
Trap: Record a conditional GO and have the bias remediated during the training runs
Why wrong: Wrong-answer trap. A conditional GO reads as a balanced compromise and is common in practice, but it carries an unresolved finding forward as governance debt and still lets unfit data into training.
Correct answer was: D
Record NO-GO on bias and have stakeholders pick the IV.4 or III.1 loop
- D4-013 · Domain IV · IV.2 — Oversee AI/ML model QA/QC documentation-required
Trap: Unit testing, code review, integration testing, and release regression checks
Why wrong: Wrong-answer trap. This is genuine software QA and reads as thorough engineering rigor, which is what makes it tempting, but it verifies code paths rather than model behavior, bias, or configuration traceability.
Correct answer was: D
Configuration management, performance verification, bias, and documentation
- D4-014 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) documentation-required
Trap: Expected levels of fit, data coverage, data volume, and pipeline reproducibility
Why wrong: Wrong-answer trap. Every item here is real CPMAI vocabulary, which is exactly what makes it convincing — but coverage, volume, and pipeline reproducibility are data-preparation questions settled at the Phase III Go/No-Go, one gate earlier and on a different artifact.
Correct answer was: B
Expected levels of fit, business KPIs, and evaluation of external and generative models
- D4-015 · Domain IV · IV.3 — Manage AI/ML model training documentation-required
Trap: The ML engineer, who owns the training infrastructure and executes the runs on it
Why wrong: Wrong-answer trap. The ML engineer really does own the training infrastructure, which makes this the most believable wrong answer, but running the platform is not the same as owning the training work itself.
Correct answer was: A
The data scientist trains; the ML engineer assists; the PM manages the work
- D4-016 · Domain IV · IV.5 — Verify data quality (GATE) iteration-trigger
Trap: Record GO and have the engineer document the workstation setup for the record
Why wrong: Wrong-answer trap. Documenting the setup is a real reproducibility practice and produces exactly the kind of artifact an audit trail wants, but a written description of one workstation is not verification on production-like infrastructure.
Correct answer was: A
Record NO-GO on reproducibility; stakeholders pick the IV.4 or III.4 loop
- D4-017 · Domain IV · IV.1 — Oversee AI/ML model technique(s) cross-domain-link
Trap: GPU sizing, storage, network egress, scaling, failover, and runbook ownership
Why wrong: Wrong-answer trap. A self-hosted model really does raise every one of these, and they belong in the V.1 deployment plan's infrastructure section — but that section provisions the choice rather than justifying it.
Correct answer was: B
Pattern fit, success criteria, Trustworthy AI, operational fit, cost, and explainability
- D4-018 · Domain IV · IV.2 — Oversee AI/ML model QA/QC cross-domain-link
Trap: Treat it as a lessons-learned item and record it for the next project to address
Why wrong: Wrong-answer trap. Lessons learned is the right home for process improvements and sounds appropriately reflective, but filing the gap there does nothing about the criterion this project still has to satisfy.
Correct answer was: C
Treat it as an I.5 accountability gap and reconstruct what stakeholders can weigh
- D4-019 · Domain IV · IV.4 — Manage data transformation to conduct data preparation iteration-trigger
Trap: Have the engineer correct the transformation rule and document the rerun for IV.5
Why wrong: Wrong-answer trap. Fixing the rule and rerunning is the natural engineering reflex, arrives properly documented for the gate, and may even be correct — but it presumes the rule is wrong rather than the data it is meeting.
Correct answer was: A
Have the data scientist and SMEs establish the cause before stakeholders choose
- D4-020 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) cross-domain-link
Trap: NO-GO — update the III.1 specification to match the data that was used
Why wrong: Wrong-answer trap. Revising III.1 is the right loop when stakeholders decide the requirements themselves were wrong, and it looks like documentation hygiene — but backdating the spec to match what was already done launders the deviation instead of evaluating it.
Correct answer was: A
NO-GO — the undocumented dataset is a III.1 deviation with I.5 exposure
- D4-021 · Domain IV · IV.1 — Oversee AI/ML model technique(s) phase-identification
Trap: Phase III — Data Preparation, where the training-ready dataset is produced
Why wrong: Wrong-answer trap. Phase III sits inside Domain IV's span, which makes it a believable answer, but data preparation finishes before a technique is selected and trained.
Correct answer was: C
Phase IV — Model Development, where the technique is chosen and trained
- D4-022 · Domain IV · IV.5 — Verify data quality (GATE) go-no-go-gate
Trap: Confirm that V.1 planning waits until the trained model is available to plan against
Why wrong: Wrong-answer trap. Waiting for the trained model before touching V.1 sounds like strict gate discipline, but the deployment plan is drafted iteratively and only finalized after IV.6.
Correct answer was: A
Confirm that V.1 planning may draft now while execution waits for the IV.6 GO
- D4-023 · Domain IV · IV.3 — Manage AI/ML model training documentation-required
Trap: Loss curves, learning rate, epochs, batch size, seed, and runtime per run
Why wrong: Wrong-answer trap. These are the fields a data scientist's experiment log really does carry and the tracker draws on them, which is what makes this feel complete — but a run-level parameter dump has no progress against plan, no data version, and no blocker history.
Correct answer was: C
Progress, hyperparameters, data version, environment, interim metrics, blockers
- D4-024 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) documentation-required
Trap: Record NO-GO and have II.8 stakeholders revise the criteria to the result achieved
Why wrong: Wrong-answer trap. Revising the documented criteria with stakeholders is a legitimate outcome and comes fully dressed in governance, which is what makes it tempting — but pre-selecting it moves the target to the result achieved instead of letting stakeholders weigh the paths.
Correct answer was: D
Record a KPI failure; stakeholders weigh more work, revision, or starting over
- D4-025 · Domain IV · IV.2 — Oversee AI/ML model QA/QC iteration-trigger
Trap: Coordinate generation of synthetic shifted data for the data scientist to test on
Why wrong: Wrong-answer trap. Synthetic shifted data is a real technique and the right recourse when real operational data cannot be obtained, but generated data reflects the team's assumptions about the shift rather than the shift itself.
Correct answer was: A
Coordinate an evaluation on operational-distribution data and document it
- D4-026 · Domain IV · IV.4 — Manage data transformation to conduct data preparation documentation-required
Trap: Accept completion now and reopen the task if the IV.5 gate surfaces problems
Why wrong: Wrong-answer trap. Letting the gate catch problems sounds like it defers properly to IV.5, but the gate evaluates prepared data against criteria — it does not verify that the preparation work itself was completed as planned.
Correct answer was: A
Accept the executed plan, closed tracker, usable dataset, and IV.5 confirmation
- D4-027 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) iteration-trigger
Trap: Have the data scientist circulate the gate write-up to stakeholders for comment
Why wrong: Wrong-answer trap. A circulated write-up produces a document and invites comment, so it feels like it satisfies the requirement, but it replaces a stakeholder decision with a broadcast.
Correct answer was: A
Have the gate review convened anyway; the documented decision is the artifact
- D4-028 · Domain IV · IV.1 — Oversee AI/ML model technique(s) iteration-trigger
Trap: Have the team move to an open-source model so the patent question falls away
Why wrong: Wrong-answer trap. Open-source feels like it sidesteps IP entirely and the instinct is common, but open-source components carry their own license terms and offer no immunity from a third party's patent.
Correct answer was: D
Have legal assess applicability, then put the resulting options to stakeholders
- D4-029 · Domain IV · IV.5 — Verify data quality (GATE) documentation-required
Trap: The gap list handed to data engineering so remediation can get under way
Why wrong: Wrong-answer trap. Handing over a gap list is the most action-oriented option and gets remediation moving, which is exactly its appeal — but without the decision document there is no record of the outcome or of the loop stakeholders chose.
Correct answer was: B
The Phase III Go/No-Go Decision with findings, gaps, target phase, and sign-off
- D4-030 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) go-no-go-gate
Trap: Issue a conditional GO and track the pending legal assessment as an open item
Why wrong: Wrong-answer trap. A conditional GO keeps the schedule alive and names the open item, which makes it feel responsible, but it authorizes operationalization against a question nobody can currently answer.
Correct answer was: A
Record NO-GO, since Trustworthy AI alignment can no longer be evidenced
- D4-031 · Domain IV · IV.4 — Manage data transformation to conduct data preparation phase-identification
Trap: Phase IV — Model Development, where the technique is selected and trained
Why wrong: Wrong-answer trap. The domain's name points squarely at model development, so mapping every one of its tasks to Phase IV is the natural error — but transformation finishes before Phase IV begins.
Correct answer was: D
Phase III — Data Preparation, where raw data becomes training-ready
- D4-032 · Domain IV · IV.3 — Manage AI/ML model training cross-domain-link
Trap: Have the data scientist use the service for internal comparisons and note it
Why wrong: Wrong-answer trap. Calling the use informal makes it feel too small to govern, which is exactly the risk — the data leaves the environment either way and nothing is documented.
Correct answer was: A
Have legal, privacy, and security weigh it against the documented privacy plan
- D4-033 · Domain IV · IV.5 — Verify data quality (GATE) iteration-trigger
Trap: Record NO-GO and have stakeholders approve sourcing more data for the model
Why wrong: Wrong-answer trap. Sourcing more data is a legitimate remedy, and routing it through stakeholders looks like exactly the right governance — but it asks stakeholders to approve an outcome the PM has already settled on. The keyed answer brings the same stakeholders in to weigh the options against each other, which is what a gate decision is.
Correct answer was: A
Record a volume failure and have stakeholders weigh the IV.1 technique loop
- D4-034 · Domain IV · IV.2 — Oversee AI/ML model QA/QC documentation-required
Trap: Throughout operations — under V.3 governance, V.4 metrics, V.7 review, and audit
Why wrong: Wrong-answer trap. That is precisely the right cadence for a model already in operation, and it is where bias does real harm — but it begins after deployment, leaving the data and the development that produced the model unmeasured.
Correct answer was: D
Throughout development — in data preparation, in QA/QC, at IV.6, and in production
- D4-035 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) documentation-required
Trap: Deployment itself — release to the production user base on the gate date
Why wrong: Wrong-answer trap. Reading GO as 'deploy now' is the most common misread of this gate, but deployment is V.2 work executed against the V.1 plan, on the schedule that plan sets.
Correct answer was: B
Operationalization work — V.1 finalization, V.2 deployment, and V.3 governance
- D4-036 · Domain IV · IV.1 — Oversee AI/ML model technique(s) documentation-required
Trap: The data scientist's reasoning narrative, captured verbatim for the project record
Why wrong: Wrong-answer trap. A thorough narrative from the data scientist reads like justification and is often well argued, but it records one role's reasoning rather than the project's decision.
Correct answer was: A
The choice, rationale, pattern and criteria fit, operational and Trustworthy AI risk
- D4-037 · Domain IV · IV.5 — Verify data quality (GATE) cross-domain-link
Trap: Re-measure bias on the held-out test set so the result rests on unseen records
Why wrong: Wrong-answer trap. Moving to the held-out test set feels like the more rigorous measurement, and on unseen records it is — but a split drawn from the same source carries exactly the same distribution problem.
Correct answer was: A
Re-measure bias on the operational distribution before the gate reaches a decision
- D4-038 · Domain IV · IV.3 — Manage AI/ML model training iteration-trigger
Trap: Establish a higher-capacity technique at IV.1 before the model reaches the gate
Why wrong: Wrong-answer trap. Raising model capacity through IV.1 is the right move against a confirmed underfit and arrives properly governed, but an early plateau can equally point to a data limitation, and nobody has established which.
Correct answer was: A
Establish the cause of the plateau with the data scientist and QA/QC before IV.6
- D4-039 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) iteration-trigger
Trap: Treat robustness as met and have the gap documented for stakeholders alongside GO
Why wrong: Wrong-answer trap. Documenting the gap for stakeholders looks like transparency and keeps the record honest, but pairing it with GO lets an open question pass as answered without anyone deciding to accept it.
Correct answer was: B
Treat robustness as unevidenced and have stakeholders choose testing or acceptance
- D4-040 · Domain IV · IV.5 — Verify data quality (GATE) documentation-required
Trap: The announcement email telling the team and stakeholders the gate outcome
Why wrong: Wrong-answer trap. An announcement reaches everyone who needs to know and feels sufficient, but a notification carries no findings and no sign-off, so it is not the gate artifact.
Correct answer was: C
The Phase III Go/No-Go Decision with findings, the GO, sign-off, and next actions
- D4-041 · Domain IV · IV.2 — Oversee AI/ML model QA/QC phase-identification
Trap: Phase V only, when the model is formally evaluated against the criteria
Why wrong: Wrong-answer trap. Phase V is where evaluation is formalized, so it looks like QA/QC's natural home, but confining QA/QC there means configuration history and interim performance were never captured during development.
Correct answer was: A
Phase IV and Phase V — continuous through development and evaluation
- D4-042 · Domain IV · IV.4 — Manage data transformation to conduct data preparation documentation-required
Trap: Encode features, scale values, split train and test, balance the class mix
Why wrong: Wrong-answer trap. This is real data work and sits right next to preparation in a practitioner's workflow, which is what makes it plausible — but feature encoding and train/test splitting are modeling-side steps that follow the IV.5 gate.
Correct answer was: C
Transform formats, reconcile inconsistencies, impute missing values, map fields
- D4-043 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) go-no-go-gate
Trap: Record NO-GO and add the baseline comparison once deployment has begun
Why wrong: Wrong-answer trap. Deferring the comparison protects the schedule and promises the evidence later, but a measure taken after deployment can no longer inform the decision to deploy.
Correct answer was: C
Record NO-GO, since the baseline comparison has not been carried out at all
- D4-044 · Domain IV · IV.1 — Oversee AI/ML model technique(s) iteration-trigger
Trap: Coordinate an evaluation on project data so the results can settle the question
Why wrong: Wrong-answer trap. Testing the model on project data produces concrete evidence and feels decisive, but performance results cannot surface licensing exposure or bias inherited from training data nobody can inspect.
Correct answer was: D
Coordinate a provenance review with legal, compliance, and security before approval
- D4-045 · Domain IV · IV.3 — Manage AI/ML model training documentation-required
Trap: Data, elapsed time, hardware, and the resources assigned to the training run
Why wrong: Wrong-answer trap. Swapping technique fit for elapsed time keeps three of the four areas and reads like a normal schedule review, which is what makes it tempting — but it drops the lens most likely to explain the overrun.
Correct answer was: C
Data, technique fit, hardware and resources, and the results produced so far
- D4-046 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) cross-domain-link
Trap: Have the GO recorded now and the privacy signature collected on the return
Why wrong: Wrong-answer trap. Promising a signature on return preserves the appearance of the requirement, but a review conducted after the decision cannot influence it.
Correct answer was: C
Have an interim privacy authority documented so the review genuinely happens
- D4-047 · Domain IV · IV.2 — Oversee AI/ML model QA/QC documentation-required
Trap: Data science, quality assurance, the data steward, and the PM, who signs off
Why wrong: Wrong-answer trap. This carries four roles and the right multi-team shape, which is what makes it convincing — but it drops the ML engineer's configuration management and security's own control validation, and it makes the PM the signatory rather than the coordinator.
Correct answer was: C
Data science, ML engineering, the data steward, and security, PM coordinating
- D4-048 · Domain IV · IV.5 — Verify data quality (GATE) pattern-recognition
Trap: Workbook steps completed with no outstanding roadblocks remaining
Why wrong: Wrong-answer trap. Workbook completion reads as project bookkeeping rather than a test of the data, so it is the item candidates most often assume does not belong here. Module 03 asks for exactly this before the project moves past Phase III: the necessary steps completed and checked off, with no roadblocks outstanding.
Correct answer was: B
Throughput of the prepared-data pipeline once it is in production
- D4-049 · Domain IV · IV.3 — Manage AI/ML model training documentation-required
Trap: The data scientist's judgment that the model is finished and performing well
Why wrong: Wrong-answer trap. The data scientist is closest to the model and their judgment carries weight, which is what makes this easy to accept, but one role's assessment is not a completion criterion the gate can act on.
Correct answer was: C
Training and validation done, results documented, QA/QC engaged, ready for IV.6
- D4-050 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) go-no-go-gate
Trap: Record NO-GO until the difference between the user segments has been eliminated
Why wrong: Wrong-answer trap. Holding out for identical performance across segments sounds like the fairest possible standard, but perfect parity is rarely achievable and the documented tolerance already expresses what stakeholders agreed to accept.
Correct answer was: D
Record GO and carry the segment differences into V.4 monitoring and V.7 contingency
- D4-051 · Domain IV · IV.4 — Manage data transformation to conduct data preparation iteration-trigger
Trap: Establish additional data-engineering capacity so the pipelines land on the plan
Why wrong: Wrong-answer trap. Adding capacity is a familiar PM lever and looks like owning the date, but it spends people before anyone knows whether effort is even the constraint.
Correct answer was: C
Establish the root cause with the team, then take the impact to stakeholders
- D4-052 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) documentation-required
Trap: The project sponsor, as the stakeholder accountable for the overall investment
Why wrong: Wrong-answer trap. The sponsor is the most senior person in the room and executives really do sign off on investments, which is what makes this feel right — but a sponsor signature alone leaves the compliance and technical questions unattested.
Correct answer was: C
The PM facilitates; stakeholders sign off, including business and security roles
- D4-053 · Domain IV · IV.1 — Oversee AI/ML model technique(s) documentation-required
Trap: Ensure every technique decision is personally approved before the team acts on it
Why wrong: Wrong-answer trap. Personal approval sounds like accountability and plenty of PMs work this way, but approving decisions you cannot evaluate is rubber-stamping, and it is not what the ECO asks the PM to do.
Correct answer was: B
Ensure the choice is documented and justified; the technical call stays with them
- D4-054 · Domain IV · IV.2 — Oversee AI/ML model QA/QC iteration-trigger
Trap: Have the ML engineer set a fixed seed, then document the rerun for the gate
Why wrong: Wrong-answer trap. Setting a seed is the standard first fix, frequently helps, and here even arrives documented — but non-determinism also comes from library versions, hardware, and data ordering that a seed does not touch.
Correct answer was: D
Have the source of the variance established, since IV.6 tests reproducibility
- D4-055 · Domain IV · IV.5 — Verify data quality (GATE) iteration-trigger
Trap: Record the removal of the correlated feature in the data preparation plan
Why wrong: Wrong-answer trap. Dropping the feature is a real mitigation and the right thing to record once stakeholders choose it, but removing a predictive feature costs performance and may not eliminate the proxy effect.
Correct answer was: C
Record the proxy as a bias finding; stakeholders weigh removal or constraints
- D4-056 · Domain IV · IV.3 — Manage AI/ML model training cross-domain-link
Trap: Assess a move to another provider with stakeholders before the maintenance begins
Why wrong: Wrong-answer trap. Moving providers looks decisive and removes the maintenance risk outright, and here it even runs through stakeholders — but it is a material infrastructure change with cost, security, and data-residency implications that need assessing first.
Correct answer was: A
Assess the interruption risk with the data scientist and cloud team, then log it
- D4-057 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) documentation-required
Trap: The rationale narrative explaining why the model was judged ready to deploy
Why wrong: Wrong-answer trap. A well-argued rationale is the most substantial-looking part of the document, which is why it can pass for the whole thing, but on its own it carries no question-by-question findings and no sign-off.
Correct answer was: C
The findings against the criteria, the decision, sign-off, rationale, next actions
- D4-058 · Domain IV · IV.4 — Manage data transformation to conduct data preparation iteration-trigger
Trap: Have stakeholders approve the recency rule so transformation work continues
Why wrong: Wrong-answer trap. Most recent wins is a defensible resolution rule, is often the right one, and here even arrives documented — but adopting it without SME validation can quietly erase real history the model needs to learn.
Correct answer was: C
Have the data SMEs assess the conflicts before stakeholders pick the rule
- D4-059 · Domain IV · IV.2 — Oversee AI/ML model QA/QC documentation-required
Trap: Code, container images, manifests, release tags, secrets, and runbooks under V.1
Why wrong: Wrong-answer trap. Teams with good delivery practice version exactly these by habit, which is what makes the list feel complete — but release artifacts describe what ships under the V.1 plan rather than what produced the model.
Correct answer was: D
Code, data, model artifacts, hyperparameters, evaluation data, and environment
- D4-060 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) iteration-trigger
Trap: Record the results restated in II.8 terms so the gate has a like-for-like read
Why wrong: Wrong-answer trap. Restating the results in II.8 terms looks like it resolves the mismatch on the spot, but it assumes the documented metrics are still what stakeholders want and are computable here — and the mismatch is the reason to test that assumption with them.
Correct answer was: B
Record NO-GO and align the II.8 metrics with stakeholders before re-testing
- D4-061 · Domain IV · IV.5 — Verify data quality (GATE) documentation-required
Trap: Yes — once the data is ready, deployment work can start alongside training
Why wrong: Wrong-answer trap. Starting deployment work while training runs sounds like sensible overlap, and V.1 planning genuinely is iterative, but deployment execution is gated at IV.6 because the model may not survive evaluation.
Correct answer was: B
No — this gate authorizes training; deployment waits on the later IV.6 gate
- D4-062 · Domain IV · IV.1 — Oversee AI/ML model technique(s) go-no-go-gate
Trap: Treat it as a V.7 incident and roll back while the team works the latency problem
Why wrong: Wrong-answer trap. Invoking V.7 and rolling back is sound immediate containment and is genuinely part of handling the incident, but on its own it neither delivers real-time performance nor addresses a technique that was mis-fitted from the start.
Correct answer was: B
Treat it as a new model needing IV.1 justification and a fresh readiness gate
- D4-063 · Domain IV · IV.3 — Manage AI/ML model training documentation-required
Trap: Have the winning configuration captured once the sweep has finished running
Why wrong: Wrong-answer trap. Recording only the winning configuration is common practice and feels like enough, but without the search space and method nobody can reproduce how that configuration was reached or judge whether the search was sound.
Correct answer was: B
Have the search space, method, and final values under configuration management
- D4-064 · Domain IV · IV.5 — Verify data quality (GATE) cross-domain-link
Trap: Record the attribute gap and proceed with the attributes the pipeline produced
Why wrong: Wrong-answer trap. Working with the attributes that came through is the pragmatic engineering answer and keeps the pipeline moving, but it quietly redefines the model's scope without anyone deciding to.
Correct answer was: D
Record a coverage failure; stakeholders weigh sourcing, revision, or a narrower model
- D4-065 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) iteration-trigger
Trap: Record GO with the evaluation questions answered and a documentation gap left
Why wrong: Wrong-answer trap. Treating an accountability gap as merely documentation, next to evaluation questions that are all answered, sounds like proportionate judgment — which is precisely the reasoning the gate exists to resist.
Correct answer was: A
Record NO-GO and complete the audit trail with I.5 stakeholders before GO
- D4-066 · Domain IV · IV.2 — Oversee AI/ML model QA/QC phase-identification
Trap: Phase VI only, after the solution is deployed and monitored
Why wrong: Wrong-answer trap. Quality monitoring genuinely does continue after deployment, which makes Phase VI sound right, but that work is V.3 governance and V.4 metrics — IV.2 is the development-side discipline that precedes it.
Correct answer was: A
Phases IV and V — continuous across development and evaluation
- D4-067 · Domain IV · IV.3 — Manage AI/ML model training iteration-trigger
Trap: Pause the run, cut the burn rate with the ML engineer, and then resume it
Why wrong: Wrong-answer trap. In-flight optimization is a real lever, sounds efficient, and is properly delegated here — but tuning the burn rate without knowing the cause may only slow the same runaway spend.
Correct answer was: D
Pause the run, establish the cost cause, and put the options to stakeholders
- D4-068 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) documentation-required
Trap: IV.6 runs after V.1 is complete and approved by the deployment stakeholders
Why wrong: Wrong-answer trap. Finishing the plan before the gate sounds orderly and sequential, but a fully approved deployment plan ahead of a GO commits the project to operating a model nobody has judged ready.
Correct answer was: B
V.1 is drafted iteratively before the gate; IV.6 GO authorizes finalization
- D4-069 · Domain IV · IV.1 — Oversee AI/ML model technique(s) cross-domain-link
Trap: Have the data scientist run a quick bias check and note the result in the file
Why wrong: Wrong-answer trap. A quick check produces a number and a file note, so it looks like the analysis happened, but a measurement that is not scoped or documented cannot support a gate decision.
Correct answer was: A
Have bias analysis coordinated on the fine-tuning data for IV.5 and IV.6 to weigh
- D4-070 · Domain IV · IV.4 — Manage data transformation to conduct data preparation documentation-required
Trap: Source profiles, known quirks, sample records, and the engineer's field notes
Why wrong: Wrong-answer trap. Engineer notes often hold the most accurate knowledge about the sources, which is what makes them feel like the real documentation, but private notes are not a governed plan anyone else can execute or audit.
Correct answer was: B
Rules, architecture, error handling, ownership, reproducibility, traceability
- D4-071 · Domain IV · IV.5 — Verify data quality (GATE) documentation-required
Trap: The data steward, who is accountable for data quality and access controls
Why wrong: Wrong-answer trap. The steward's accountability for data quality makes them look like the natural owner of a data-quality gate, but their role here is to supply validation evidence, not to run the decision.
Correct answer was: C
The project manager, who convenes the review and documents the decision
- D4-072 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) iteration-trigger
Trap: Re-evaluate once the ML engineer has adapted the model to the new environment
Why wrong: Wrong-answer trap. Adapting the model first looks efficient and may even be the eventual answer, but it settles the question the gate exists to decide and leaves the change undocumented.
Correct answer was: D
Re-evaluate operational fit against the new environment before the gate decides
- D4-073 · Domain IV · IV.3 — Manage AI/ML model training documentation-required
Trap: When a model can no longer be improved — a point that does not arrive
Why wrong: Wrong-answer trap. CPMAI genuinely is iterative and models genuinely can always be improved, which makes this sound methodologically sophisticated — but iteration runs between defined completion points, and without them no gate could ever convene.
Correct answer was: B
When training and validation are done, documented, and QA/QC engaged
- D4-074 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) documentation-required
Trap: The model returns the same prediction for the same input at inference
Why wrong: Wrong-answer trap. Deterministic inference is a genuine form of repeatability and is easy to demonstrate, which is why teams offer it at the gate — but it only shows the finished model is stable, not that it could ever be rebuilt.
Correct answer was: C
The training run can be repeated from documented inputs to an equivalent model
- D4-075 · Domain IV · IV.5 — Verify data quality (GATE) go-no-go-gate
Trap: GO can be withdrawn by the PM later, with the change documented afterwards
Why wrong: Wrong-answer trap. This correctly senses that decisions are revisitable, which makes it feel close to right, but it relocates the authority to the PM — reversal comes through a documented stakeholder decision at a gate, not a unilateral withdrawal.
Correct answer was: D
GO authorizes training; later findings can still send the project back to iterate
- D5-001 · Domain V · V.1 — Manage creation of AI solution deployment plan oversight-vs-doing
Trap: Coordinate a staging validation run and engage stakeholders on the plan in parallel
Why wrong: Wrong-answer trap. A coordinated staging validation run with stakeholders engaged is the right move during Domain IV model testing, and the consultation makes it look properly governed. This model has already passed that point, so it starts deployment activity ahead of the plan meant to govern it.
Correct answer was: A
Coordinate creation of the deployment plan and obtain stakeholder sign-off first
- D5-002 · Domain V · V.4 — Oversee AI solution metrics iteration-trigger
Trap: Log the drop as within-tolerance variance and reconfirm the trend at the next review
Why wrong: Wrong-answer trap. Treating a single reading as variance and reconfirming next cycle is right when a metric dips inside its tolerance band. Eight points sustained across three months is a trend, not noise, and the wait leaves a degraded model serving.
Correct answer was: B
Establish whether drift, decay, or scope shift caused it, then engage stakeholders
- D5-003 · Domain V · V.2 — Manage AI solution deployment documentation-required
Trap: Confirm the runtime is healthy, monitoring is live, and the rollback path was tested
Why wrong: Wrong-answer trap. Runtime, live monitoring, and a tested rollback is the right closure checklist for the V.7 contingency test, and it reads as a thorough deployment check. It leaves out the performance baseline and the governance arrangements V.2 also requires.
Correct answer was: A
Confirm plan executed, monitoring live, baseline met, and governance in place
- D5-004 · Domain V · V.3 — Oversee model governance escalation-trigger
Trap: Roll back to the last authorized version and notify stakeholders of the restoration
Why wrong: Wrong-answer trap. Rollback with stakeholder notification is the correct first move when the unauthorized version is known to be misbehaving, and it looks like a decisive restoration of control. This one has served for a month with no reported defect, so validation comes before disturbing production.
Correct answer was: A
Validate the running version, document the deviation, escalate, reinforce control
- D5-005 · Domain V · V.7 — Oversee AI solution contingency plan documentation-required
Trap: Have operations author incident runbooks once production reveals the real failure modes
Why wrong: Wrong-answer trap. Operations authoring runbooks from real incident experience is the right pattern for a solution already in steady-state operation under V.6. Applied before go-live it means the first real incident happens with no documented response.
Correct answer was: A
Create and test the plan now, documenting scenarios, owners, and escalation paths
- D5-006 · Domain V · V.6 — Manage AI solution transition plan documentation-required
Trap: Notify operations and have the two teams coordinate and document the handoff
Why wrong: Wrong-answer trap. Having the two teams coordinate and document the handoff between themselves is normal delegation and works once a transition plan exists to structure it. On its own it produces a handoff shaped by whoever is in the room, with no readiness sign-off.
Correct answer was: A
Coordinate a written transition plan with knowledge transfer and sign-off
- D5-007 · Domain V · V.5 — Prepare final report / lessons learned pattern-recognition
Trap: They close the project record once filed, which is what the audit requirement needs
Why wrong: Wrong-answer trap. Filing for audit is a genuine records obligation and fully satisfies it, which is what makes the operations lead's suggestion sound complete. It closes the file without closing the iterative loop that gives lessons learned their purpose.
Correct answer was: B
Lessons learned are input to the next iteration's Phase I business understanding
- D5-008 · Domain V · V.3 — Oversee model governance cross-domain-link
Trap: Re-run the prediction on the current model and provide the reproduced result to legal
Why wrong: Wrong-answer trap. Reproducing a prediction is the right check during a V.3 re-evaluation of model reproducibility, and it looks like direct evidence. Current inputs and the current model version do not reconstruct the decision context that existed at the time.
Correct answer was: A
Provide the audit trail: input, model version, prediction, timestamp, overrides
- D5-009 · Domain V · V.4 — Oversee AI solution metrics documentation-required
Trap: Model accuracy, data quality, and drift, the three highest-signal model indicators
Why wrong: Wrong-answer trap. Accuracy, data quality, and drift is the correct set for a model-health review inside the data science team, which is what makes it feel like the AI-specific answer. It omits the business outcome the project was funded to deliver.
Correct answer was: B
Business KPIs, model performance metrics, and operational health indicators
- D5-010 · Domain V · V.1 — Manage creation of AI solution deployment plan documentation-required
Trap: Hosting provider, operations owner, cutover window, escalation, sign-off record
Why wrong: Wrong-answer trap. This is a competent standard-IT rollout checklist and it even carries escalation and sign-off, which is exactly why it tempts. It omits the serving method and the day-one monitoring commitment, the two AI-specific elements V.1 turns on.
Correct answer was: C
How served, where operated, performance, escalation, monitoring, and sign-off
- D5-011 · Domain V · V.2 — Manage AI solution deployment go-no-go-gate
Trap: Convene stakeholders to accept the monitoring gap as a documented residual risk
Why wrong: Wrong-answer trap. A stakeholder-accepted, documented residual risk is the right mechanism for an exposure the plan never committed to closing, and convening stakeholders wears the governed costume. Monitoring from day one is an explicit V.1 commitment and a V.2 completion criterion, so there is nothing here to accept.
Correct answer was: B
Pause the deployment until monitoring is live, then resume the cutover as planned
- D5-012 · Domain V · V.3 — Oversee model governance documentation-required
Trap: Access control, provenance, audit logs, version control, bias, and model retraining
Why wrong: Wrong-answer trap. Five of the six elements here are correct, which is what makes it the closest decoy. Retraining is a V.4 life-cycle response to drift, and it displaces the sharing and extension control that governs how others may build on the model.
Correct answer was: B
Access control, provenance, audit logs, version control, bias, and extension
- D5-013 · Domain V · V.4 — Oversee AI solution metrics iteration-trigger
Trap: Document the technical KPI results and declare the deployment successful in the report
Why wrong: Wrong-answer trap. Documenting green technical KPIs is real evidence and would justify declaring model performance acceptable at a V.2 completion check. V.4 measures against the II.8 business success criteria, and those are flat.
Correct answer was: B
Surface the disconnect and engage stakeholders on why predictions are not acted on
- D5-014 · Domain V · V.7 — Oversee AI solution contingency plan escalation-trigger
Trap: Convene the bias review board to assess the output and recommend remediation
Why wrong: Wrong-answer trap. A bias review board is the right forum for a fairness finding surfaced by routine V.3 monitoring, and convening it looks properly governed. A live discriminatory output affecting a real customer is an incident, and reviewing while it keeps serving leaves the harm running.
Correct answer was: D
Execute the documented incident procedure: contain the output, then investigate and notify
- D5-015 · Domain V · V.5 — Prepare final report / lessons learned documentation-required
Trap: Coordinate operations' confirmation and document that they can operate it
Why wrong: Wrong-answer trap. A coordinated, documented confirmation from the receiving team is exactly the V.6 criterion for transition being complete, which makes it sound like the closing condition. Closeout is broader than transition and ends in stakeholder sign-off on the final report.
Correct answer was: A
Convene closeout: final report, lessons learned, transition decision, sign-off
- D5-016 · Domain V · V.1 — Manage creation of AI solution deployment plan phase-identification
Trap: Phase I — Business Understanding, where deployment goals are set
Why wrong: Wrong-answer trap. Phase I does set the success criteria Domain V measures against, but it maps to Domain II business understanding.
Correct answer was: B
Phase VI — Operationalization, where the model runs in production
- D5-017 · Domain V · V.2 — Manage AI solution deployment pattern-recognition
Trap: Real-time prediction, where the model answers each request as it arrives
Why wrong: Wrong-answer trap. Real-time prediction is easy to confuse with stream learning, and candidates often assume only one of the pair made PMI's list. Both are on it: real-time serves each request, while stream learning also updates the model continuously.
Correct answer was: C
Edge inference, where the model runs on the device capturing the data
- D5-018 · Domain V · V.3 — Oversee model governance cross-domain-link
Trap: Convene a model review with stakeholders, compare segments, and deprecate one
Why wrong: Wrong-answer trap. A stakeholder model review that compares segments and consolidates to one version is the right call when a single version dominates everywhere, and the consultation wears the governed costume. Here each version serves a segment better, so deprecating either discards demonstrated value for one group.
Correct answer was: B
Convene a governance review of routing logic, fairness, and accountability
- D5-019 · Domain V · V.4 — Oversee AI solution metrics documentation-required
Trap: Yes — a sustained decline at three months indicates the deployment was flawed
Why wrong: Wrong-answer trap. A sustained decline does warrant investigation, and attributing it to deployment execution is the instinctive read a sponsor will reach for. It treats an anticipated condition as a defect.
Correct answer was: C
No — drift is inevitable, which is why monitoring and response were planned
- D5-020 · Domain V · V.6 — Manage AI solution transition plan iteration-trigger
Trap: Coordinate an indefinite SME escalation path and complete the transition on schedule
Why wrong: Wrong-answer trap. A coordinated SME escalation path is genuinely required by V.6, so this looks like meeting the requirement properly. Escalation covers exceptional situations; it does not substitute for routine operating capability, and indefinite dependency means transition never completes.
Correct answer was: C
Pause transition and coordinate skill-gap remediation before ownership moves
- D5-021 · Domain V · V.5 — Prepare final report / lessons learned documentation-required
Trap: Deployment plan, sign-off record, and the change log covering the release
Why wrong: Wrong-answer trap. The deployment plan with its sign-off and change log is a rigorous and complete artifact set, which is exactly why it tempts. It is the V.1 and V.3 record of what was authorized before deployment, not a report on the outcome.
Correct answer was: A
Summary, performance against criteria, lessons learned, recommendations, risks
- D5-022 · Domain V · V.7 — Oversee AI solution contingency plan pattern-recognition
Trap: Performance breach, with automated rollback or a pause on serving
Why wrong: Wrong-answer trap. A performance breach looks like a V.4 monitoring finding rather than a contingency, which is why candidates rule it out. The documented rollback or serving pause makes it a core V.7 scenario.
Correct answer was: B
A new regulation taking effect that adds explainability obligations
- D5-023 · Domain V · V.1 — Manage creation of AI solution deployment plan cross-domain-link
Trap: Log an optimization work item and have engineering fit the model to CPU-only
Why wrong: Wrong-answer trap. A logged optimization item is the right engineering path once stakeholders have chosen cost over capability, and it is what most teams would try first. Chosen here it converts a technique-versus-cost decision into an engineering default.
Correct answer was: D
Surface the operational-fit conflict and take the resolution to stakeholders
- D5-024 · Domain V · V.4 — Oversee AI solution metrics iteration-trigger
Trap: Schedule documented capture improvements, assign an owner, and record progress
Why wrong: Wrong-answer trap. A documented improvement plan with a named owner and tracked progress is the right cadence for a platform capability enhancement, and the record-keeping makes it look properly governed. Traceability is an I.5 accountability control, so incremental delivery means untraceable decisions keep accumulating.
Correct answer was: D
Treat it as a governance gap, document it, and remediate before further reliance
- D5-025 · Domain V · V.3 — Oversee model governance documentation-required
Trap: Document the request in the model registry and have engineering share the weights
Why wrong: Wrong-answer trap. Registering the request and having engineering fulfil it is the right flow for an internal team consuming the model under existing access controls, and the registry entry looks like proper governance. An external party fine-tuning it raises licensing, accountability, and IP questions extension controls exist to answer.
Correct answer was: B
Run the request through V.3 extension controls with legal and compliance
- D5-026 · Domain V · V.2 — Manage AI solution deployment iteration-trigger
Trap: Document the finding on the risk register and patch it in the next scheduled release
Why wrong: Wrong-answer trap. A risk-register entry plus next-release patching is the correct handling for a low-severity finding security has already triaged, and it wears the documented costume. Nobody has assessed severity here, so the acceptance rests on nothing.
Correct answer was: A
Pause and have security assess severity before remediating or accepting risk
- D5-027 · Domain V · V.4 — Oversee AI solution metrics documentation-required
Trap: Quarterly, aligning metric review with the documented business reporting cycle
Why wrong: Wrong-answer trap. A quarterly rhythm is right for the business-value reporting that goes to a steering committee, and anchoring it to a documented organizational cycle sounds disciplined. Model metrics need a tighter loop, because drift can degrade an AI solution well inside a quarter.
Correct answer was: D
At the cadence documented in the deployment plan, sized to metric criticality
- D5-028 · Domain V · V.7 — Oversee AI solution contingency plan documentation-required
Trap: Yes — a documented and approved contingency plan satisfies the requirement
Why wrong: Wrong-answer trap. Documented and approved is the bar for most planning artifacts, and it would be the right answer for the deployment plan or the transition plan. V.7 specifically adds the testing requirement on top.
Correct answer was: A
No — the plan must be tested before production to demonstrate it works
- D5-029 · Domain V · V.1 — Manage creation of AI solution deployment plan go-no-go-gate
Trap: Deploy under a documented performance exception and close the gap post-go-live
Why wrong: Wrong-answer trap. A documented performance exception is a legitimate governance instrument and shipping under one is common practice, which is what makes it tempting. It knowingly deploys a solution that cannot meet the requirement stakeholders approved.
Correct answer was: B
Loop back to the operational-fit gate and take the options to stakeholders
- D5-030 · Domain V · V.3 — Oversee model governance iteration-trigger
Trap: Add bias monitoring to the next release and document the gap in the governance log
Why wrong: Wrong-answer trap. Bundling instrumentation into the next release with a governance-log entry is orderly roadmap and records discipline, so it reads as properly governed. It extends unmeasured bias exposure by a release cycle and leaves the six months already served unexamined.
Correct answer was: C
Add bias measurement now and assess the six months of predictions already served
- D5-031 · Domain V · V.5 — Prepare final report / lessons learned cross-domain-link
Trap: Have finance prepare the variance analysis, brief the sponsor, and file it
Why wrong: Wrong-answer trap. Finance producing the variance analysis and briefing the sponsor is sound collaboration and often how the numbers get built, which is what makes the delegation feel correct. The PM still owns documenting the variance and its causes in the report itself.
Correct answer was: B
Compare actual versus projected ROI, analyze the causes, engage stakeholders
- D5-032 · Domain V · V.6 — Manage AI solution transition plan documentation-required
Trap: Production credentials, service accounts, and the runtime configuration set
Why wrong: Wrong-answer trap. Credentials, service accounts, and runtime configuration are genuinely required and are the first thing any receiving team asks for. Access without runbooks, governance, and contingency is not operating capability.
Correct answer was: C
Model artifacts, governance plan, monitoring, contingency, runbooks, contacts
- D5-033 · Domain V · V.4 — Oversee AI solution metrics cross-domain-link
Trap: Approve the per-demographic calibration and record the rationale in the audit trail
Why wrong: Wrong-answer trap. The intent is genuinely corrective, the technique is defensible, and recording the rationale in the audit trail looks like proper governance. Differential treatment by demographic needs legal and ethics review before it ships, not documentation after.
Correct answer was: A
Treat it as a bias incident and take the remediation options to governance review
- D5-034 · Domain V · V.1 — Manage creation of AI solution deployment plan documentation-required
Trap: The ML engineer, the platform lead, and the release manager who execute it
Why wrong: Wrong-answer trap. Engineer, platform lead, and release manager is the right approval set for a change-control record inside the runtime they own, and they are closest to the work. Executing is not authorizing, which is the oversight-versus-doing distinction V.1 tests.
Correct answer was: B
Sponsor, platform and security, operations, compliance, and business owner
- D5-035 · Domain V · V.2 — Manage AI solution deployment oversight-vs-doing
Trap: Approve the change under the plan's emergency provision and record it afterwards
Why wrong: Wrong-answer trap. An emergency provision with after-the-fact recording is the right path when production is down and the fix cannot wait, and invoking a documented provision looks properly governed. This deployment is proceeding normally, so schedule pressure is not an emergency.
Correct answer was: A
Pause and route the architecture change through the plan's change-review process
- D5-036 · Domain V · V.7 — Oversee AI solution contingency plan iteration-trigger
Trap: Record the test failure in the contingency log and proceed with the deployment
Why wrong: Wrong-answer trap. Logging the failed test is exactly what a disciplined PM does and the contingency log is its right home. Logging it and proceeding leaves production with no working recovery path.
Correct answer was: C
Halt deployment, have the rollback procedure fixed, and re-test before go-live
- D5-037 · Domain V · V.3 — Oversee model governance documentation-required
Trap: A/B comparative testing between candidate versions serving live traffic
Why wrong: Wrong-answer trap. A/B testing reads as an experimentation practice rather than a control, which is why candidates rule it out. PMI places it inside the governance framework for multi-version production.
Correct answer was: D
Inference cost allocation across the business units consuming the model
- D5-038 · Domain V · V.4 — Oversee AI solution metrics documentation-required
Trap: Published industry benchmarks for comparable models in the same sector
Why wrong: Wrong-answer trap. Industry benchmarks are legitimate context, often appear in the V.5 report, and carry an air of authority. They are not what this project committed to deliver.
Correct answer was: C
The Domain II success criteria together with the model evaluation results
- D5-039 · Domain V · V.6 — Manage AI solution transition plan phase-identification
Trap: At close-out, when the final report, handover, and sign-off are completed
Why wrong: Wrong-answer trap. Close-out is the most common trigger, is the one most PMs picture, and the transition usually does execute there. It is one trigger among several rather than the rule.
Correct answer was: C
Whenever a handoff is triggered: team change, vendor change, successor project
- D5-040 · Domain V · V.5 — Prepare final report / lessons learned iteration-trigger
Trap: Document the underestimate in lessons learned and close the final report
Why wrong: Wrong-answer trap. Recording the observation meets the documentation requirement and is what most closeouts do, which is why it feels sufficient. V.5 lessons exist to change the next iteration, and an observation with no recommendation attached changes nothing.
Correct answer was: D
Document it with specific recommendations for the next iteration's ROI method
- D5-041 · Domain V · V.1 — Manage creation of AI solution deployment plan iteration-trigger
Trap: Document a manual operating procedure and run the model under that procedure
Why wrong: Wrong-answer trap. A documented manual operating procedure is a familiar and often workable pattern, and writing it down looks properly governed. It commits the organization to running an AI solution with no drift detection and no automated rollback.
Correct answer was: A
Treat it as an operational-fit gap and resolve it with stakeholders before GO
- D5-042 · Domain V · V.4 — Oversee AI solution metrics documentation-required
Trap: Technical KPIs are the diagnostic layer; business KPIs are what stakeholders act on
Why wrong: Wrong-answer trap. Technical metrics are the diagnostic layer and stakeholders do act on business numbers, so the framing is half right. It casts technical KPIs as subordinate when V.4 requires both as first-class oversight signal.
Correct answer was: A
Business KPIs measure outcomes, technical KPIs measure behavior, both required
- D5-043 · Domain V · V.7 — Oversee AI solution contingency plan cross-domain-link
Trap: Escalate to the security team, which owns breach response across the organization
Why wrong: Wrong-answer trap. Security genuinely leads the technical breach response, so escalating to them reads as correct role clarity. The PM still owns customer notification, the regulatory obligations, and the AI-specific remediation.
Correct answer was: C
Execute the documented contingency plan, beginning with containment of the breach
- D5-044 · Domain V · V.3 — Oversee model governance documentation-required
Trap: No — governance belongs to the data steward who controls access and lineage
Why wrong: Wrong-answer trap. The data steward does own access and lineage controls and is a necessary participant, so this sounds like correct delegation to the right specialist. Governance authority sits with the PM-overseen program spanning engineering, security, legal, and compliance.
Correct answer was: C
No — the PM oversees governance while execution spans several functions
- D5-045 · Domain V · V.2 — Manage AI solution deployment documentation-required
Trap: Accept the engineer's report and mark the deployment task complete in the plan
Why wrong: Wrong-answer trap. Accepting the report of the person who did the work is normal, usually right, and the report is valid evidence for one criterion. Deployed and deployment complete are different claims under V.2.
Correct answer was: C
Verify against the documented success criteria and record the verification
- D5-046 · Domain V · V.4 — Oversee AI solution metrics iteration-trigger
Trap: Reduce the frequency and coordinate the update to the plan's documented cadence
Why wrong: Wrong-answer trap. Coordinating the update to the plan's documented cadence is exactly the right bookkeeping once a change is agreed, so this wears the governed costume convincingly. It records a risk decision the PM took alone instead of one stakeholders made.
Correct answer was: B
Take the cost-versus-detection tradeoff to stakeholders and document it
- D5-047 · Domain V · V.6 — Manage AI solution transition plan cross-domain-link
Trap: Note that malicious-AI response sits with the organization's security function
Why wrong: Wrong-answer trap. Security does lead breach response organizationally, so writing that into the transition plan reads as correct role clarity. The AI-specific detection, containment, and response procedures still transition with the solution.
Correct answer was: D
Transfer the malicious-AI contingency plan and train the team to execute it
- D5-048 · Domain V · V.5 — Prepare final report / lessons learned documentation-required
Trap: Include the raw audit trail so stakeholders can verify the report's findings
Why wrong: Wrong-answer trap. Full transparency to stakeholders is usually the right instinct and is what an open project culture rewards. Distributing raw audit records defeats the access control that makes them evidentiary under I.5.
Correct answer was: D
Decline, and route the stakeholder to the audit system through governance
- D5-049 · Domain V · V.3 — Oversee model governance iteration-trigger
Trap: Decommission the two unauthorized versions, log the change, and restore baseline
Why wrong: Wrong-answer trap. Decommissioning with a logged change and a restored baseline is decisive, is right when a version is known to be causing harm, and looks properly governed. Done before the impact assessment it can break dependent systems and destroy the evidence of how they got there.
Correct answer was: C
Investigate origin, impact, and whether the gap is systemic, then remediate
- D5-050 · Domain V · V.7 — Oversee AI solution contingency plan documentation-required
Trap: After each incident, so the plan is revised against what actually happened
Why wrong: Wrong-answer trap. Post-incident revision is genuinely required, is a mature practice, and V.7 does expect it. A schedule driven only by incidents means the plan is only ever tested by real failures.
Correct answer was: A
At the cadence in the plan, and whenever a major change makes it stale
- D5-051 · Domain V · V.1 — Manage creation of AI solution deployment plan pattern-recognition
Trap: Model operationalization, where the model serves predictions in production
Why wrong: Wrong-answer trap. Operationalization is where the model runs in production and is easy to conflate with integration, but it covers serving and running rather than wiring the model into an app.
Correct answer was: C
Model scaffolding, the integration layer between the model and consuming apps
- D5-052 · Domain V · V.4 — Oversee AI solution metrics cross-domain-link
Trap: Add the benchmark next release, record the gap, and set a remediation date
Why wrong: Wrong-answer trap. A recorded gap with a committed remediation date is orderly roadmap and governance discipline, and it is the right handling for an enhancement. It still extends the period in which a committed fairness criterion goes unmeasured and leaves past predictions unassessed.
Correct answer was: B
Add the metric now, assess past predictions, and document the gap and fix
- D5-053 · Domain V · V.3 — Oversee model governance cross-domain-link
Trap: Update the model card and have data science add post-hoc output explanations
Why wrong: Wrong-answer trap. Post-hoc explainability is the natural technical response, may well end up part of the answer, and updating the model card is proper governance hygiene. Committing to it before legal establishes the required standard risks building to the wrong bar.
Correct answer was: B
Establish the requirement with legal, then weigh explainability against a change
- D5-054 · Domain V · V.2 — Manage AI solution deployment iteration-trigger
Trap: Have the engineer add input validation to reject the malformed requests
Why wrong: Wrong-answer trap. Boundary validation is the most likely eventual fix, which is exactly what makes jumping to it risky: if the malformation starts upstream, validation hides a data pipeline defect instead of surfacing it.
Correct answer was: C
Investigate where the malformed input originates before deciding the fix
- D5-055 · Domain V · V.6 — Manage AI solution transition plan documentation-required
Trap: Only partly — independent operation plus documented escalation paths
Why wrong: Wrong-answer trap. Documented escalation paths are genuinely one of the V.6 transition-plan artifacts, so this looks like the right second condition. They are part of what gets handed over, but it is the receiving team's sign-off that closes the transition.
Correct answer was: D
Only partly — independent operation plus documented artifacts and receiving-team sign-off
- D5-056 · Domain V · V.5 — Prepare final report / lessons learned documentation-required
Trap: That the sponsor assign a more experienced project manager to the next iteration
Why wrong: Wrong-answer trap. Routing a staffing recommendation through the sponsor sounds like a decisive fix escalated to the right authority. Attributing a process gap to individual capability produces no transferable practice and no organizational learning.
Correct answer was: D
A communication plan with cadence, audiences, and escalation criteria defined
- D5-057 · Domain V · V.4 — Oversee AI solution metrics iteration-trigger
Trap: Have the data scientist retrain the model on the most recent production data
Why wrong: Wrong-answer trap. Retraining on recent data is the standard drift response and would be exactly right if the model were drifting. The analysis says it is not, so retraining refits the model to a target the business has already redefined.
Correct answer was: B
Take it to stakeholders as a business-needs question, not a model question
- D5-058 · Domain V · V.7 — Oversee AI solution contingency plan pattern-recognition
Trap: A dedicated ethics board that reviews every AI decision the system makes in production
Why wrong: Wrong-answer trap. An ethics board is a real governance mechanism and reads as the most rigorous option on offer, which is what makes it tempting. Reviewing decisions after they are made is not a human in the decision-making loop, and it is not what PMI names as the foundation.
Correct answer was: B
Human oversight — keep a human in the loop and represent the AI system with a human
- D5-059 · Domain V · V.3 — Oversee model governance documentation-required
Trap: The user identifier, the session context, and the model version invoked
Why wrong: Wrong-answer trap. Identity and session context are genuinely logged and feel like the accountability record, and they are the right fields for an access-control audit. They capture who asked rather than what the model decided and why.
Correct answer was: D
Input, model version, prediction, timestamp, rationale, and overrides
- D5-060 · Domain V · V.1 — Manage creation of AI solution deployment plan iteration-trigger
Trap: Have engineering test caching, batching, and quantization to close the gap
Why wrong: Wrong-answer trap. Caching, batching, and quantization are legitimate latency optimizations and are how most teams would start, so this reads as competent engineering management. An undocumented workaround substitutes for a cost-versus-scope decision stakeholders need to make.
Correct answer was: D
Put the tradeoff to stakeholders with the scope, cost, and technique options
- D5-061 · Domain V · V.2 — Manage AI solution deployment documentation-required
Trap: Have the change documented while the deployment continues on its current path
Why wrong: Wrong-answer trap. Documenting a change while work continues is the right pattern for a low-impact fix inside an already-approved change window, so it looks like the governed compromise. It records a deviation after the fact instead of authorizing it beforehand.
Correct answer was: C
Pause and put the proposed change through plan revision and sign-off
- D5-062 · Domain V · V.6 — Manage AI solution transition plan iteration-trigger
Trap: Transition as planned and log the gaps on the receiving team's improvement backlog
Why wrong: Wrong-answer trap. A backlog entry looks like an agreed plan and is the normal way engineering debt moves between teams. It hands over known I.5 exposure with no acceptance from the receiving team and no date attached.
Correct answer was: C
Close the gaps first, or transition under a documented remediation plan
- D5-063 · Domain V · V.5 — Prepare final report / lessons learned documentation-required
Trap: Focus on failures, since the practices that worked will be repeated anyway
Why wrong: Wrong-answer trap. Successful practices do tend to persist inside one continuing team, which makes this sound like an efficient use of the closeout session. They are exactly what is lost when the team changes.
Correct answer was: C
Capture both — successful practices transfer as much as failures do
- D5-064 · Domain V · V.4 — Oversee AI solution metrics documentation-required
Trap: Nothing — usage data is too privacy-sensitive to collect on a live model
Why wrong: Wrong-answer trap. Privacy sensitivity is a real constraint and would be the right answer if the question were about retaining raw user content. It shapes the collection controls rather than removing a monitoring requirement.
Correct answer was: C
Who uses the model, how, and how often, feeding governance and versions
- D5-065 · Domain V · V.7 — Oversee AI solution contingency plan iteration-trigger
Trap: Continue serving while retraining proceeds against recent production data
Why wrong: Wrong-answer trap. Keeping the service up while retraining is the least disruptive option and is often correct once the impact is known to be small. Chosen first, nobody has judged what the degraded predictions are costing.
Correct answer was: D
Assess impact, then decide with stakeholders whether a rollback is needed
- D5-066 · Domain V · V.3 — Oversee model governance phase-identification
Trap: They are the same requirement, tagged to two domains for exam coverage
Why wrong: Wrong-answer trap. The overlap is real enough that they read as one requirement, and both produce audit records, which is what makes this tempting. They sit at different levels: one states the obligation, the other implements it.
Correct answer was: A
I.5 sets the accountability requirement; V.3 enforces it in production
- D5-067 · Domain V · V.1 — Manage creation of AI solution deployment plan documentation-required
Trap: After the model-readiness gate authorizes the move to operationalization
Why wrong: Wrong-answer trap. The IV.6 gate is genuinely when the plan is finalized and signed, which makes this half right and hard to reject. Starting the draft there forfeits the plan's influence on technique and architecture.
Correct answer was: A
Early, refined through the project, and finalized after the readiness gate
- D5-068 · Domain V · V.2 — Manage AI solution deployment cross-domain-link
Trap: Move the hardening into the next release and document the plan deviation
Why wrong: Wrong-answer trap. Sequencing hardening into the next release with a documented deviation is ordinary release management and looks properly recorded. It puts production inference data on an environment the organization has already judged inadequate.
Correct answer was: A
Pause and have the environment hardened to standard before resuming
- D5-069 · Domain V · V.6 — Manage AI solution transition plan pattern-recognition
Trap: A successor project replacing the solution now in production
Why wrong: Wrong-answer trap. A successor project can look like ordinary lifecycle progression, but it hands the solution to a different team and requires the documented transition.
Correct answer was: A
A scheduled retraining cycle producing an updated model version
- D5-070 · Domain V · V.5 — Prepare final report / lessons learned iteration-trigger
Trap: Omit it from the report, since the contingency plan resolved the situation
Why wrong: Wrong-answer trap. A resolved risk feels like a closed item, and it is closed in the register. The successful response is precisely the reusable lesson V.5 exists to carry into the next iteration.
Correct answer was: C
Document the risk, what occurred, the response, and what to reuse
- D5-071 · Domain V · V.4 — Oversee AI solution metrics documentation-required
Trap: Restricted access, with the data scientist reviewing, curating, and relaying
Why wrong: Wrong-answer trap. A single expert curating the numbers looks like quality control and is the right arrangement when metrics are genuinely access-restricted. It removes the direct stakeholder visibility the dashboard exists to provide.
Correct answer was: A
Defined access, review cadence, alerting, and traceability to criteria
- D5-072 · Domain V · V.7 — Oversee AI solution contingency plan documentation-required
Trap: Security and legal, who lead incident response and regulatory notification
Why wrong: Wrong-answer trap. Security and legal do lead organizational incident response and regulatory notification, which makes them a plausible pair to own the exercise. The AI-specific rollback and the stakeholder decision points sit outside their remit.
Correct answer was: A
Engineering, operations, security, stakeholders, and the receiving team
- D5-073 · Domain V · V.3 — Oversee model governance iteration-trigger
Trap: Document the performance record as evidence that re-evaluation is not due
Why wrong: Wrong-answer trap. Performance evidence is real and would properly close out a V.4 metrics review, so documenting it looks rigorous. It does not cover bias drift, reproducibility, or slow distribution shift, which is what re-evaluation examines.
Correct answer was: A
Schedule a documented re-evaluation, since performance is one dimension
- D5-074 · Domain V · V.1 — Manage creation of AI solution deployment plan cross-domain-link
Trap: Anonymize the inference data before transfer so no personal data crosses
Why wrong: Wrong-answer trap. Anonymization is a legitimate privacy control, is the most technically appealing answer here, and would resolve a consent-scope conflict. Many jurisdictional rules govern data location regardless of identifiability, so it needs legal confirmation rather than assumption.
Correct answer was: A
Escalate the conflict to privacy and legal and resolve it before sign-off
- D5-075 · Domain V · V.2 — Manage AI solution deployment documentation-required
Trap: Mark it successful, since error-free serving was the cutover exit condition
Why wrong: Wrong-answer trap. Clean serving is the most visible signal of a good cutover and is a valid exit condition for the cutover step itself, which is exactly why PMI tests whether you stop there. V.2 completion also covers monitoring, baseline, and governance.
Correct answer was: D
Verify runtime, monitoring, performance, and governance before declaring
- CV2-D3-001 · Domain III · III.1 — Define required data oversight-vs-doing
Trap: Coordinating the workspace, infrastructure, and tooling the team will use to stage gathered data
Why wrong: Wrong-answer trap. Coordinating workspace and infrastructure is ECO III.4 and is correct once required data is defined and sources are known. It answers 'where will the team work,' not 'what data does the project need.'
Correct answer was: A
Specifying the attributes, sources, and aggregation needed to meet the Phase I success criteria
- CV2-D3-002 · Domain III · III.2 — Identify data SMEs oversight-vs-doing
Trap: Log the ambiguity in the risk register and resolve it at the III.8 data gate with stakeholders
Why wrong: Wrong-answer trap. Carrying an unresolved data question to the III.8 gate is correct when the question genuinely cannot be answered before then. This one is answerable now, and the gate is weeks away — modeling would proceed on guessed semantics in the meantime.
Correct answer was: A
Identify current data stewards or domain experts who can clarify the unclear field semantics
- CV2-D3-003 · Domain III · III.3 — Identify data sources and locations cross-domain-link
Trap: Coordinate with legal and security to compress the vendor review, then re-baseline the schedule
Why wrong: Wrong-answer trap. Working legal and security to shorten the review is correct follow-through once stakeholders have chosen to wait for the vendor data. Doing it first quietly makes that choice for them and commits the schedule to an option they never saw.
Correct answer was: C
Document the 12-week dependency as a project risk and engage stakeholders on the trade-off
- CV2-D3-004 · Domain III · III.3 — Identify data sources and locations documentation-required
Trap: Type, volume, access mechanism, refresh cadence, legal terms, endpoint, and ingest owner
Why wrong: Wrong-answer trap. This set is nearly right — it swaps accuracy for volume and cost for ingest ownership. Accuracy is what III.7 evaluation measures against and source cost feeds the project's business case, so neither can be traded for a staffing or sizing detail.
Correct answer was: C
Type, cost, access mechanism, accuracy, refresh cadence, legal terms, and endpoint
- CV2-D3-005 · Domain III · III.5 — Gather required data iteration-trigger
Trap: Convene the III.8 Go/No-Go, present the current gap, and let stakeholders weigh iterating back
Why wrong: Wrong-answer trap. Convening III.8 is correct when gathering is complete and the residual gap is fully characterized. Gathering is still underway and no alternatives have been assessed, so the Go/No-Go would be asked to decide on information the team has not produced yet.
Correct answer was: B
Loop back to ECO III.3 to evaluate alternative sources and re-assess the impact on the III.8 gate
- CV2-D3-006 · Domain III · III.6 — Check data privacy, compliance, and access cross-domain-link
Trap: Coordinate with the data protection officer to anonymize the dataset, then resume analysis
Why wrong: Wrong-answer trap. Coordinating anonymization with the data protection officer is the right control when the project is minimizing data for a lawful, in-scope purpose — and involving that officer makes it look governed. It does not cure a purpose-limitation breach that has already occurred, and whether it could is a legal determination, not a PM one.
Correct answer was: A
Treat as a compliance incident: pause use, escalate to legal, do not proceed unilaterally
- CV2-D3-007 · Domain III · III.7 — Oversee data evaluation oversight-vs-doing
Trap: Convene the III.8 gate and let stakeholders question the evaluation's completeness there
Why wrong: Wrong-answer trap. Convening the gate and letting stakeholders probe the evidence is right once the evaluation package has been verified complete. ECO III.7 puts that completeness check on the PM first, so handing an unverified package to the gate outsources the PM's own task to the people relying on it.
Correct answer was: A
Verify the evaluation covered the documented data quality criteria before authorizing
- CV2-D3-008 · Domain III · III.8 — Determine if data meets solution needs go-no-go-gate
Trap: Convene stakeholders, authorize a return to III.3, and source the missing 20% of attributes
Why wrong: Wrong-answer trap. Returning to III.3 is the correct path when the missing attributes are known to exist in a source nobody has examined yet. The stem says they are unavailable from ANY identified source, so iteration is a hypothesis stakeholders must weigh against cost — not a conclusion the PM can hand them pre-decided.
Correct answer was: A
Convene the documented III.8 Go/No-Go with stakeholders and record the decision and basis
- CV2-D3-009 · Domain III · III.8 — Determine if data meets solution needs escalation-trigger
Trap: Record the skew as an I.3 finding and require it to be re-measured and decided at the IV.6 gate
Why wrong: Wrong-answer trap. Logging a bias measurement as a tracked finding and re-deciding it at IV.6 is right when the disparity sits inside the Trustworthy AI tolerance set in Phase I and only needs monitoring. A 40% underrepresentation is outside any such tolerance, and III.8 asks whether the data is fit for use — a promise to look again later does not answer that.
Correct answer was: A
Escalate the bias finding now: it touches I.3 and the III.8 gate cannot pass without a decision
- CV2-D3-010 · Domain III · III.7 — Oversee data evaluation iteration-trigger
Trap: Convene the III.8 Go/No-Go again so stakeholders can re-decide against the shifted data
Why wrong: Wrong-answer trap. Re-opening the gate is the right escalation once re-characterization shows the shift invalidates the basis of the GO. Convening stakeholders before anyone can tell them what actually changed spends their time and produces a decision with no evidence behind it.
Correct answer was: C
Re-characterize the variance, update the III.8 gate documentation, and engage stakeholders
- CV2-D4-001 · Domain IV · IV.1 — Oversee AI/ML model technique(s) (e.g., algorithm, selection) oversight-vs-doing
Trap: Coordinate a parallel build of both model classes and bring the accuracy comparison to compliance
Why wrong: Wrong-answer trap. A parallel build is the right investment when compliance's open question is empirical: how much accuracy an interpretable model actually costs. Here the open question is whether a black-box model is permissible at all, so this answers a different question at considerable expense.
Correct answer was: C
Document the trade-off and engage compliance and stakeholders for the technique decision
- CV2-D4-002 · Domain IV · IV.2 — Oversee AI/ML model QA/QC (e.g., configuration management, model performance) documentation-required
Trap: Coordinate with the data engineer to reconstruct the configurations from the server logs
Why wrong: Wrong-answer trap. Reconstructing configurations from logs is the correct move when one specific past run must be reproduced — a regulator's question, or a IV.6 baseline comparison. As a general remedy it is archaeology: hyperparameter choices made in local notebooks were never written to a server log, so the reconstruction would be confidently incomplete.
Correct answer was: C
Require all configuration changes to be recorded in a documented change log going forward
- CV2-D4-003 · Domain IV · IV.3 — Manage AI/ML model training iteration-trigger
Trap: Convene stakeholders, re-baseline the schedule, and document the revised training estimate
Why wrong: Wrong-answer trap. Re-baselining with stakeholders is the correct step once root cause is understood and the revised estimate rests on something. Re-baselining an unexplained overrun asks stakeholders to approve a number nobody can defend, and it typically buys a second overrun with their signature on it.
Correct answer was: D
Pause and conduct a structured root-cause review across data, technique, hardware, results
- CV2-D4-004 · Domain IV · IV.4 — Manage data transformation to conduct data preparation oversight-vs-doing
Trap: Convene stakeholders to re-confirm the II.8 success criteria before preparation is accepted
Why wrong: Wrong-answer trap. Re-confirming II.8 success criteria with stakeholders is the right move when a sponsor has signalled a priority shift, as at the IV.6 gate. Nothing in the stem says the criteria moved, so re-opening them convenes the right people about the wrong question and delays the gate that actually blocks training.
Correct answer was: C
Verify the IV.5 prepared-data gate is convened and its decision documented before training
- CV2-D4-005 · Domain IV · IV.5 — Verify data quality for go/no-go decision to conduct data preparation go-no-go-gate
Trap: Convene stakeholders, rescope the project, and reset the II.8 completeness target to 84%
Why wrong: Wrong-answer trap. This convenes stakeholders and documents the outcome, which is what gate discipline looks like — except it arrives with the outcome already chosen. Rescoping is a remedy the project reaches after a No-Go sends it back to an earlier phase, not a proposal the PM carries into the gate in place of the decision itself.
Correct answer was: D
Convene stakeholders for the IV.5 Go/No-Go and document the decision against the II.8 criteria
- CV2-D4-006 · Domain IV · IV.5 — Verify data quality for go/no-go decision to conduct data preparation escalation-trigger
Trap: Record a GO with the disparity as a documented I.3 risk and re-measure it at the IV.6 gate
Why wrong: Wrong-answer trap. Logging a disparity as a tracked risk and re-measuring at IV.6 is the right handling when the measurement sits inside the Phase I tolerance. This one exceeds it, and IV.5 evaluates bias on prepared data specifically so unfit data does not reach training in the first place.
Correct answer was: A
Treat as a cross-domain gate failure: engage I.3 stakeholders to decide mitigation path
- CV2-D4-007 · Domain IV · IV.6 — Verify model ready for operationalization go/no-go decision go-no-go-gate
Trap: Record a GO against the II.8 criteria, document the gap, and log robustness as a V.4 watch item
Why wrong: Wrong-answer trap. Passing on performance and tracking robustness in production is right when robustness was never among the documented IV.6 criteria for this solution. It is one of them here, and moving an unevaluated criterion into V.4 monitoring means the first robustness test runs against live users.
Correct answer was: C
Convene stakeholders to assess the robustness gap and make a documented Go/No-Go call
- CV2-D4-008 · Domain IV · IV.6 — Verify model ready for operationalization go/no-go decision cross-domain-link
Trap: Convene stakeholders to update the II.8 success criteria before the gate is scheduled
Why wrong: Wrong-answer trap. Convening stakeholders to rewrite II.8 wears full governance dress and reaches the right artifact, but it skips the one conversation that determines whether the rewrite is needed. The sponsor owns the success criteria, so whether the shift belongs in this release is theirs to say before anyone is convened.
Correct answer was: B
Surface the priority shift, decide with sponsor whether to update II.8 before the gate
- CV2-D4-009 · Domain IV · IV.6 — Verify model ready for operationalization go/no-go decision documentation-required
Trap: Convene the gate and let stakeholders request the baseline numbers they consider material
Why wrong: Wrong-answer trap. Letting stakeholders drive what evidence they want is right at a gate whose documented criteria are already satisfied and where the question is depth, not completeness. A baseline comparison IS one of the IV.6 criteria, so convening without it hands the PM's completeness check to the people who depend on it.
Correct answer was: D
Request the specific baseline metrics and comparison numbers before convening the IV.6 gate
- CV2-D4-010 · Domain IV · IV.6 — Verify model ready for operationalization go/no-go decision oversight-vs-doing
Trap: Convene a second stakeholder review to re-confirm the criteria before documenting the GO
Why wrong: Wrong-answer trap. A second review is the right call when a criterion was contested or the evidence was thin the first time. Every criterion here is documented and met, and manufacturing another cycle is the mirror image of skipping the gate — governance theatre rather than governance.
Correct answer was: D
Document the GO decision with all stakeholders present, then transition to V.1 with handoff
- CV2-D5-001 · Domain V · V.1 — Manage creation of AI solution deployment plan oversight-vs-doing
Trap: Authorize a limited pilot deployment and complete the deployment plan before full rollout
Why wrong: Wrong-answer trap. A limited pilot is the right opening move when the deployment plan already exists and names the pilot as its first stage — that is staged rollout, not shortcut. With no plan drafted, a pilot is deployment without the performance criteria, escalation paths, and monitoring that V.1 exists to establish.
Correct answer was: C
Coordinate creation of the deployment plan with stakeholder sign-off before deployment
- CV2-D5-002 · Domain V · V.1 — Manage creation of AI solution deployment plan documentation-required
Trap: Serving mechanism, hosting platform, expected user load, on-call rotation, and vendor list
Why wrong: Wrong-answer trap. Hosting platform, user load, on-call rotation, and a vendor list are exactly what a standard IT rollout plan captures, and they are the right set for one. V.1 asks for the AI-specific elements — performance criteria traced to II.8 and a monitoring approach for model behavior — none of which appear here.
Correct answer was: A
Serving mechanism, operation location, performance criteria, escalation, monitoring, and sign-off
- CV2-D5-003 · Domain V · V.2 — Manage AI solution deployment documentation-required
Trap: Confirm plan executed, runtime healthy, sponsor sign-off obtained, then declare complete
Why wrong: Wrong-answer trap. Sponsor sign-off reads as the strongest possible confirmation, and it is the right gate — for the V.1 plan, before deployment. At V.2 the question is whether the plan was executed and the production baseline is met, and a signature establishes neither.
Correct answer was: D
Confirm plan executed, monitoring live, performance baseline met, governance in place
- CV2-D5-004 · Domain V · V.3 — Oversee model governance escalation-trigger
Trap: Document the change retroactively, update the change log, and reinforce the process going forward
Why wrong: Wrong-answer trap. Back-filling the change log is the right correction when a change genuinely went through review and only the record was missed. No review happened here, so the entry would document a version nobody has validated against requirements — bookkeeping standing in for incident response.
Correct answer was: A
Treat as governance incident: validate version, document deviation, escalate, reinforce
- CV2-D5-005 · Domain V · V.4 — Oversee AI solution metrics (e.g., KPI, model performance) iteration-trigger
Trap: Coordinate an immediate retraining cycle, log it as a V.3 change, and notify the sponsor
Why wrong: Wrong-answer trap. Coordinating a retraining cycle and recording it as a governed change is the right action once triage points to model decay and fresh labeled data exists. If the cause is upstream data drift or a shift in how the model is being used, retraining reproduces the same failure on a new schedule and consumes the budget that would have diagnosed it.
Correct answer was: D
Investigate root cause across drift, model decay, scope shift; engage stakeholders on decision
- CV2-D5-006 · Domain V · V.4 — Oversee AI solution metrics (e.g., KPI, model performance) documentation-required
Trap: Business KPIs, model performance, and cost per inference tracked on three separate views
Why wrong: Wrong-answer trap. Two of the three categories are right and cost per inference is a legitimate operational metric, which makes this the closest miss. The failure is 'three separate views': oversight decisions come from reading business, model, and operational signals against each other, and siloed dashboards make that impossible.
Correct answer was: C
Business KPIs, model performance, and operational health together on one shared view
- CV2-D5-007 · Domain V · V.4 — Oversee AI solution metrics (e.g., KPI, model performance) oversight-vs-doing
Trap: Coordinate with the data scientist to compile a current model performance summary for the sponsor
Why wrong: Wrong-answer trap. Asking the data scientist for a current performance summary is right when the sponsor's question is technical — accuracy, latency, error rates. Business value is measured against II.8 success criteria, and V.4 makes tracking it the PM's own standing responsibility rather than something to source on request.
Correct answer was: B
Assemble the V.4 monitoring metrics into a current benefits view against the II.8 criteria
- CV2-D5-008 · Domain V · V.7 — Oversee AI solution contingency plan escalation-trigger
Trap: Coordinate a manual rollback runbook with the ML engineer as the documented interim procedure
Why wrong: Wrong-answer trap. A documented manual procedure is a reasonable interim control when an automated rollback is degraded but the manual path has itself been proven. Nothing here has ever been executed under production conditions, so this records an untested procedure as a control — the same gap in better packaging.
Correct answer was: B
Halt the deployment until rollback is fixed and re-tested under realistic production conditions
- CV2-D5-009 · Domain V · V.7 — Oversee AI solution contingency plan iteration-trigger
Trap: Invoke the rollback procedure the contingency plan defines for this class of incident
Why wrong: Wrong-answer trap. Immediate rollback is exactly right when the documented criteria are plainly met — that is what pre-authorized contingency is for, and hesitating would be the error. The stem's problem is that the team cannot tell whether the criteria are met, and a V.7 plan supplies criteria to apply, not a reflex to execute.
Correct answer was: A
Convene stakeholders on impact, decide rollback per documented criteria, capture lessons
- CV2-D5-010 · Domain V · V.3 — Oversee model governance documentation-required
Trap: Re-run the prediction on the logged inputs and supply the reproduction as the decision record
Why wrong: Wrong-answer trap. A reproduction run is useful corroboration when the trail is already complete and the regulator wants the result verified. Standing alone it proves nothing about the original decision — the model version in force, the inputs as they stood, and any human override at that moment are exactly what the trail captures and a re-run cannot recover.
Correct answer was: C
Provide the audit trail: input, model version, prediction, timestamp, rationale, overrides
- CV2-D1-001 · Domain I · I.1 — Oversee privacy and security plan cross-domain-link
Trap: GDPR data processing agreement, regulatory framework documentation, and access controls
Why wrong: Wrong-answer trap. A data processing agreement is genuinely the correct instrument, under GDPR, for personal data from EU residents — the structure of the answer is right and only the jurisdiction is wrong. The stem places the project and the data in US healthcare, where HIPAA governs.
Correct answer was: B
HIPAA Business Associate Agreement, regulatory framework documentation, and access controls
- CV2-D1-002 · Domain I · I.1 — Oversee privacy and security plan oversight-vs-doing
Trap: Sign off on the complete sections and track incident response as an open action item
Why wrong: Wrong-answer trap. Partial sign-off with a tracked action is right for a living artifact whose sections stand on their own — a phased test plan, say — and it keeps the team moving. A privacy plan is the authorization to handle regulated data, so signing it while its breach-response half is missing approves handling with no answer for the worst case.
Correct answer was: D
Return the plan to the team with a requirement to add incident response before any sign-off
- CV2-D1-003 · Domain I · I.2 — Manage AI/ML transparency (e.g., data selection, algorithm selection) cross-domain-link
Trap: Approve the approach and record the post-hoc explanation tooling in the I.2 transparency log
Why wrong: Wrong-answer trap. Recording the explanation tooling in the transparency documentation is exactly right once compliance has accepted post-hoc explanation for this risk tier, and the documentation step makes it look governed. Post-hoc explanations satisfy some regulators and not others, and deciding they satisfy the EU AI Act is a compliance interpretation the PM cannot make.
Correct answer was: B
Surface the trade-off to compliance and stakeholders before technique selection is locked
- CV2-D1-004 · Domain I · I.3 — Conduct bias checks (e.g., model, data, algorithm) documentation-required
Trap: Convene stakeholders to mandate pre-processing data rebalancing as the project mitigation
Why wrong: Wrong-answer trap. This mandates a different mitigation class and convenes stakeholders to do it, which is why it reads as the governed option. Rebalancing training data is right when the disparity has been established as data-sourced, and even then the choice is the stakeholders' to make rather than the PM's to hand them.
Correct answer was: A
Document the finding, present mitigation options, engage stakeholders, ensure monitoring
- CV2-D1-005 · Domain I · I.3 — Conduct bias checks (e.g., model, data, algorithm) phase-identification
Trap: At the IV.5 prepared-data gate and again at IV.6, where the model is evaluated
Why wrong: Wrong-answer trap. IV.5 and IV.6 are both real bias checkpoints, and naming two of them reads as more complete than naming one. Both sit downstream of the data work — by IV.5 the skewed data has already been prepared, and by IV.6 it has already shaped the model.
Correct answer was: A
Starting in Phase II during data understanding and continuing through all later phases
- CV2-D1-006 · Domain I · I.4 — Monitor regulatory and policy compliance iteration-trigger
Trap: Coordinate with compliance, add the required disclosure language, and update the V.1 deployment plan
Why wrong: Wrong-answer trap. Adding disclosure language with compliance is precisely right once the impact assessment shows disclosure is the only affected area — right partner, right artifact, right change. Disclosure laws routinely reach further into consent flow, opt-out handling, and audit logging, and fixing the named symptom first tends to end the assessment before it starts.
Correct answer was: A
Document the regulatory change, assess impact across domains, engage stakeholders on path
- CV2-D1-007 · Domain I · I.4 — Monitor regulatory and policy compliance cross-domain-link
Trap: Coordinate with legal to add a GDPR data processing agreement for the EU data sources
Why wrong: Wrong-answer trap. A data processing agreement is the correct contractual instrument for EU personal data and legal is the right partner to draft it, which makes this look like the compliant move. Signing one does not make the project GDPR-ready: I.1 plan elements, III.6 consent and purpose limitation, and the III.8 gate all shift when the jurisdiction does, and proceeding on that basis presumes an assessment nobody has performed.
Correct answer was: B
Update the privacy plan to cover GDPR, surface impact to stakeholders, reassess III.8 gate
- CV2-D1-008 · Domain I · I.5 — Manage accountability documentation and audit trail documentation-required
Trap: Model version, prediction value, timestamp, decision rationale, validation report, sign-off
Why wrong: Wrong-answer trap. Validation report and sign-off are genuine accountability artifacts — from the IV.6 gate, where the MODEL was approved rather than where this decision was made. The set also drops the input, so nothing here shows what the model was actually reacting to.
Correct answer was: B
Input data, model version, prediction value, timestamp, decision rationale, override records
- CV2-D1-009 · Domain I · I.5 — Manage accountability documentation and audit trail oversight-vs-doing
Trap: Convene stakeholders to agree which decision types the audit logging policy should cover
Why wrong: Wrong-answer trap. Convening stakeholders to set a coverage policy reads as the governed answer, and it is the right conversation for retention windows or sampling of non-regulated telemetry. Audit coverage is an I.5 baseline rather than a policy question — the gaps a selective-coverage policy produces are exactly the ones regulators ask about.
Correct answer was: D
Require audit trail logging to be operational at deployment as part of the V.1 plan execution
- CV2-D1-010 · Domain I · I.5 — Manage accountability documentation and audit trail cross-domain-link
Trap: Convene the data scientist and operations, reconstruct the missing rationale, and release it
Why wrong: Wrong-answer trap. It convenes the right people and produces the missing content, which is why it reads as the thorough option — and reconstructing the reasoning is genuinely useful for fixing the logging. Offered in response to a discrimination claim it is manufactured evidence, and considerably worse than the gap if anyone later identifies it as reconstructed.
Correct answer was: B
Acknowledge the rationale gap, escalate as a governance finding, provide what exists, fix
- CV2-D2-001 · Domain II · II.1 — Identify problem to be solved (e.g., needs, persona) phase-identification
Trap: Convene stakeholders to define measurable success criteria for the customer service solution
Why wrong: Wrong-answer trap. Facilitating measurable success criteria is real PM work under ECO II.8 and it convenes exactly the right people, which is what makes it tempting. Criteria measure the achievement of a defined problem; written before the problem is named, they end up measuring an ambition.
Correct answer was: B
Identify the specific customer service problem and which Three Ps of Intelligence apply
- CV2-D2-002 · Domain II · II.2 — Evaluate initial AI feasibility phase-identification
Trap: Coordinate a proof of concept to validate data availability before the feasibility review convenes
Why wrong: Wrong-answer trap. A proof of concept is the right investment when data viability is the specific dimension in doubt and a cheap test would settle it. It covers one leg of a three-legged assessment, so running it first substitutes evidence on one dimension for the feasibility review that weighs all three.
Correct answer was: A
Convene a feasibility review across business viability, data viability, and technology readiness
- CV2-D2-003 · Domain II · II.6 — Manage adoption/integration risks documentation-required
Trap: Convene the sponsor and end users, then bring compliance and legal in at the II.9 business case
Why wrong: Wrong-answer trap. Sponsor and end users first is a defensible sequence and II.9 is a real place where compliance weighs in, which makes the deferral sound planned rather than skipped. Holding compliance and legal until the business case means the requirements they own land after the project has already committed to an approach.
Correct answer was: D
Engage the business sponsor, end users, compliance, legal, and operations alongside engineering
- CV2-D2-004 · Domain II · II.4 — Develop AI project scope statement oversight-vs-doing
Trap: Document accuracy as the II.8 success criterion and confirm it with the sponsor in writing
Why wrong: Wrong-answer trap. Documenting a success criterion and obtaining written sponsor confirmation is real II.8 discipline and produces a signed artifact, which is why it reads as the rigorous option. What it signs is the same unmeasurable word: success criteria are derived FROM elicited requirements, so this locks the gap in rather than closing it.
Correct answer was: C
Facilitate workshops with stakeholders to elicit functional and non-functional requirements
- CV2-D2-005 · Domain II · II.7 — Draft AI solution phase-identification
Trap: Incremental (staged delivery)
Why wrong: Wrong-answer trap. Incremental is the closest miss because staged delivery reads like iteration. Incremental builds pieces of a known solution in sequence; adaptive revisits the solution itself each cycle, which is what discovering data quality and model behavior requires.
Correct answer was: C
Adaptive/agile (short cycles)
- CV2-D2-006 · Domain II · II.10 — Identify project resources (e.g., people, hardware, contractors) oversight-vs-doing
Trap: Convene stakeholders to re-baseline the schedule around the data scientist's domain ramp-up
Why wrong: Wrong-answer trap. Re-baselining with stakeholders is the governed way to absorb a ramp-up, and it is the right move once the project has decided to buy the learning curve. ECO II.10 asks the PM to close the skills gap in the resource structure first; paying for ramp-up is what you do when no SME is available.
Correct answer was: C
Pair a domain SME with the data scientist and document the gap in the role definitions
- CV2-D2-007 · Domain II · II.5 — Determine ROI documentation-required
Trap: Model retraining compute, monitoring tooling, and an incident response retainer for operations
Why wrong: Wrong-answer trap. Every item here is real and genuinely post-deployment, which makes it look like the operational category the estimate is missing. It is one slice of that category, and it says nothing about acquiring the data or paying for third-party services and labeling — the costs that most often double an AI estimate.
Correct answer was: C
Data acquisition costs, third-party services, and the ongoing operational costs post-deployment
- CV2-D2-008 · Domain II · II.8 — Define success criteria (e.g., KPIs, metrics) documentation-required
Trap: Document the sponsor's statement and have the data scientist convert it to a measurable target
Why wrong: Wrong-answer trap. Recording what the sponsor said and having the practitioner make it measurable looks like a sensible division of labour, and it is the right split when the sponsor's statement already carries a number to formalize. 'Most customer queries' carries none, so the data scientist would be choosing the business target — which ECO II.8 assigns to stakeholders.
Correct answer was: B
Facilitate definition of measurable thresholds, business KPIs, and bias/fairness boundaries
- CV2-D2-009 · Domain II · II.3 — Conduct risk assessment(s) (e.g., security, safety, ethics) documentation-required
Trap: Data risks, regulatory risks, technology risks, vendor risks, and information security exposure
Why wrong: Wrong-answer trap. Four of these five are right, and vendor and security exposure are genuine concerns on AI projects, which makes this the closest miss. Both sit inside the technology and regulatory categories, and the set drops the two with no traditional-IT analogue: ethical risk, and whether the organization will actually adopt the output.
Correct answer was: C
Data risks, regulatory risks, ethical risks, technology risks, and organizational adoption risks
- CV2-D2-010 · Domain II · II.4 — Develop AI project scope statement documentation-required
Trap: Document the sponsor's comprehensive scope and manage delivery through a release backlog
Why wrong: Wrong-answer trap. Managing a broad scope through a release backlog is exactly right on an adaptive project — it is how iterative delivery absorbs discovery, and it wears the correct methodology. It presumes an agreed scope to sequence, and 'comprehensive' is an ambition, so a backlog built from it hides the boundary decision instead of making it.
Correct answer was: B
Facilitate scope definition: transaction types, fraud patterns, and release boundaries