Wrong-Answer Trap Library
256 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
Why wrong: Wrong-answer trap — re-running doesn't reproduce original context.
Correct answer was: B
Provide the audit trail from accountability documentation: input, model version, prediction, timestamp, decision rationale, human-in-the-loop overrides — sourced from the documented program (I.5)
- D1-002 · Domain I · I.3 — Conduct bias checks documentation-required
Trap: Approve the post-processing
Why wrong: Wrong-answer trap — post-processing without governance.
Correct answer was: B
Document the finding, evaluate mitigation options (pre-processing data balance, in-processing fairness constraints, post-processing calibration), engage stakeholders for the decision, ensure mitigation continues into production monitoring (V.4)
- D1-003 · Domain I · I.4 — Monitor regulatory and policy compliance phase-identification
Trap: False — only Phase I
Why wrong: Wrong-answer trap — Phase I-only is too narrow.
Correct answer was: B
False — Domain I runs continuously throughout all six phases. Privacy plans live from Phase I through VI. Bias checks happen in Phases II/III/IV/V. Compliance monitoring is continuous.
- D1-004 · Domain I · I.2 — Manage AI/ML transparency cross-domain-link
Trap: Add post-hoc XAI after deployment
Why wrong: Wrong-answer trap — post-hoc XAI may not satisfy regulator.
Correct answer was: B
Document the trade-off between performance and explainability for stakeholders; consider interpretable-by-design alternatives; engage compliance/legal early; do not approve unilaterally without surfacing the constraint
- D1-005 · Domain I · I.1 — Oversee privacy and security plan cross-domain-link
Trap: Have the engineer proceed with anonymization in parallel
Why wrong: Wrong-answer trap — parallel work bypasses governance.
Correct answer was: B
Treat as I.1 + III.6 cross-pull: coordinate with privacy/legal, validate anonymization meets regulatory bar (re-identification risk assessed), document the decision, update privacy plan; do not approve unilaterally
- D1-006 · Domain I · I.2 — Manage AI/ML transparency documentation-required
Trap: Decision transparency requires open-source
Why wrong: Wrong-answer trap — open-source is one approach to systemic, not required.
Correct answer was: B
Systemic transparency = visibility into the model's construction (data, preprocessing, architecture, training). Decision transparency = visibility into specific predictions (why was THIS decision made). Systemic is achievable; decision is harder.
- D1-007 · Domain I · I.3 — Conduct bias checks pattern-recognition
Trap: All three are equally relevant
Why wrong: Wrong-answer trap — only Informational matters for fairness.
Correct answer was: C
I — Informational bias (over/underrepresentation in data with fairness implications)
- D1-009 · Domain I · I.4 — Monitor regulatory and policy compliance cross-domain-link
Trap: Have the data scientist add post-hoc explanations
Why wrong: Wrong-answer trap — XAI may not satisfy regulator's bar.
Correct answer was: B
Coordinate with legal/compliance to assess regulation requirements; engage data science to add XAI capability OR consider model replacement (loop IV.1) if XAI insufficient; update governance plan; engage stakeholders on schedule/scope/cost impact
- D1-010 · Domain I · I.1 — Oversee privacy and security plan documentation-required
Trap: Permission, Re-identification, Identification, Disclosure, Encryption
Why wrong: Wrong-answer trap — different mapping.
Correct answer was: A
Personal data identified, Regulatory frameworks, Incident response, Data access controls, Encryption/anonymization
- D1-011 · Domain I · I.2 — Manage AI/ML transparency documentation-required
Trap: They're the same
Why wrong: Wrong-answer trap — they're distinct.
Correct answer was: A
Disclosure = users know AI made the decision; Consent = users can opt out with feasible alternatives
- D1-012 · Domain I · I.3 — Conduct bias checks documentation-required
Trap: Identify, Eliminate, Verify
Why wrong: Wrong-answer trap — sounds plausible but wrong vocab.
Correct answer was: A
Pre-processing (rebalance data), In-processing (training-time fairness constraints), Post-processing (output calibration)
- D1-013 · Domain I · I.5 — Manage accountability documentation and audit trail cross-domain-link
Trap: False — audit trails aren't required
Why wrong: Wrong-answer trap — audit trails are required.
Correct answer was: B
False — audit trails are stored securely with controlled access per governance plan; only authorized auditors and per-need-to-know access; retention per regulatory requirements
- D1-014 · Domain I · I.4 — Monitor regulatory and policy compliance documentation-required
Trap: CCPA
Why wrong: Wrong-answer trap — CCPA is California.
Correct answer was: B
GDPR
- D1-015 · Domain I · I.1 — Oversee privacy and security plan cross-domain-link
Trap: I.2 (transparency) only
Why wrong: Wrong-answer trap — transparency is part but not the full picture.
Correct answer was: B
I.1 (HIPAA privacy), I.3 (health data has known bias risks), I.4 (HIPAA, possibly state laws), I.5 (health decisions need named human accountability)
- D1-016 · Domain I · I.2 — Manage AI/ML transparency documentation-required
Trap: True — feedback-driven is best practice
Why wrong: Wrong-answer trap — feedback-driven sounds good but wrong here.
Correct answer was: B
False — disclosure is upfront, not retrofit. Deploying a chatbot without telling users it's AI is a Trustworthy AI violation that user feedback shouldn't have to surface.
- D1-017 · Domain I · I.3 — Conduct bias checks iteration-trigger
Trap: Approve per-demographic calibration
Why wrong: Wrong-answer trap — well-meaning but governance bypass.
Correct answer was: B
Recognize as bias incident (V.4 + I.3 + I.4 cross-pull): document; engage privacy/legal/ethics; evaluate options (re-source, retrain with rebalanced data, post-processing fairness, descope); do not unilaterally apply per-demographic adjustment without governance
- D1-018 · Domain I · I.5 — Manage accountability documentation and audit trail documentation-required
Trap: Audit trail is a subset of systemic transparency
Why wrong: Wrong-answer trap — they don't subset.
Correct answer was: B
Systemic transparency = visibility into model construction (data, architecture, training); part of I.2. Audit trail = chronological record of decisions/uses; part of I.5. Different artifacts; both required.
- D1-019 · Domain I · I.4 — Monitor regulatory and policy compliance iteration-trigger
Trap: Have legal handle it without project involvement
Why wrong: Wrong-answer trap — legal-only without project context.
Correct answer was: B
Coordinate with legal/compliance to assess applicability and timeline; engage stakeholders if compliance requires project changes (scope, timeline, technique); update compliance map; do not assume regulations don't apply
- D1-020 · Domain I · I.1 — Oversee privacy and security plan go-no-go-gate
Trap: Conditional GO with privacy plan pending
Why wrong: Wrong-answer trap — conditional GO with pending governance.
Correct answer was: B
ITERATE — III.8 includes trustworthy-AI alignment (I.1 cross-pull). Document the privacy plan gap; do not GO until privacy plan sign-off is complete.
- D1-021 · Domain I · I.3 — Conduct bias checks documentation-required
Trap: True — but only if regulators require
Why wrong: Wrong-answer trap — methodology principle independent of regulation.
Correct answer was: A
True
- D1-022 · Domain I · I.2 — Manage AI/ML transparency documentation-required
Trap: Interpretability is post-hoc; XAI is inherent
Why wrong: Wrong-answer trap — reverses the definitions.
Correct answer was: B
XAI = post-hoc methods to explain decisions of any model. Interpretability = inherently understandable models (e.g., decision trees). High-stakes typically prefers interpretability.
- D1-023 · Domain I · I.5 — Manage accountability documentation and audit trail documentation-required
Trap: The model itself
Why wrong: Wrong-answer trap — model can't be accountable.
Correct answer was: B
A human must be named as accountable for consequential AI decisions; this doesn't mean a human reviews every prediction, but for any consequential decision, a human is named responsible — tier by stakes (high-stakes = HITL, medium = HOTL, low = automated with framework above)
- D1-024 · Domain I · I.4 — Monitor regulatory and policy compliance phase-identification
Trap: Phase IV only
Why wrong: Wrong-answer trap — Phase IV-only is partial.
Correct answer was: B
Throughout — Phase I scope assessment, Phase II data sourcing/privacy, Phase IV governance, Phase VI production. Continuous monitoring per I.4.
- D1-025 · Domain I · I.1 — Oversee privacy and security plan iteration-trigger
Trap: Continue serving production while patching
Why wrong: Wrong-answer trap — known vulnerability serving = governance violation.
Correct answer was: B
Execute the contingency response (V.7): assess severity (data exposure, ongoing risk), contain (patch, restrict access, or pause), audit (was data compromised?), notify (per regulatory and customer-protection requirements), remediate, document; coordinate with security/legal/leadership
- D1-026 · Domain I · I.2 — Manage AI/ML transparency cross-domain-link
Trap: Approve since users may prefer human-feeling experience
Why wrong: Wrong-answer trap — user 'preference' doesn't override.
Correct answer was: B
Block deployment: disclosure is a Trustworthy AI Framework requirement (I.2); users should be told they're interacting with AI; surface to stakeholders and compliance for resolution
- D1-027 · Domain I · I.3 — Conduct bias checks iteration-trigger
Trap: Have the data scientist add a disclaimer
Why wrong: Wrong-answer trap — disclaimer doesn't fix bias.
Correct answer was: B
Recognize that 'reflecting real-world differences' isn't a defense if the AI amplifies or perpetuates the disparity in ways the success criteria didn't intend; engage ethics/legal/stakeholders to evaluate whether disparity is acceptable per use case; document the decision; possibly loop back to II.8 to clarify success criteria around fairness
- D1-028 · Domain I · I.5 — Manage accountability documentation and audit trail cross-domain-link
Trap: I.4 only
Why wrong: Wrong-answer trap — regulatory alone is partial.
Correct answer was: B
All five — V.3 enforces all Domain I principles in production: privacy plan (I.1), transparency (I.2), bias monitoring (I.3), regulatory compliance (I.4), accountability/audit (I.5)
- D1-029 · Domain I · I.4 — Monitor regulatory and policy compliance documentation-required
Trap: True — but only for executives
Why wrong: Wrong-answer trap — narrow scope.
Correct answer was: A
True — workforce trained in AI governance is foundational; PM advocates and tracks training as compliance work
- D1-030 · Domain I · I.2 — Manage AI/ML transparency documentation-required
Trap: Regulators can contest AI deployment
Why wrong: Wrong-answer trap — regulatory contestability is part but not the user-facing principle.
Correct answer was: B
Users affected by AI decisions can challenge them: disclosure (they know AI made decision) + explanation (XAI) + mechanism to challenge + human review path + outcome remediation. Required under EU AI Act and GDPR Article 22.
- D2-001 · Domain II · II.1 — Identify problem to be solved oversight-vs-doing
Trap: Approve the project as scoped
Why wrong: Wrong-answer trap — vague scope can't be approved.
Correct answer was: B
Engage stakeholders to define the specific problem to be solved (II.1) — need, persona, business outcome — before any solution is drafted; vague goals like 'improve customer experience' are insufficient as project basis
- D2-002 · Domain II · II.8 — Define success criteria documentation-required
Trap: Have the data scientist define metrics
Why wrong: Wrong-answer trap — data scientist defines technical metrics, not business success.
Correct answer was: B
Facilitate stakeholder engagement to define measurable, time-bound, business-tied success criteria (CSAT score uplift target, baseline, measurement methodology, time horizon); vague success criteria fail downstream evaluation gates (III.7, III.8, IV.6, V.4)
- D2-003 · Domain II · II.5 — Determine ROI documentation-required
Trap: Skip Trustworthy AI costs since they're not material
Why wrong: Wrong-answer trap — assumed-immaterial without analysis.
Correct answer was: B
Coordinate with finance/governance to include Trustworthy AI costs (Domain I): privacy plan ops, audit infrastructure, bias monitoring tools, governance review cadence, contestability mechanism; engage stakeholders if the more accurate ROI changes the project's value proposition
- D2-004 · Domain II · II.7 — Draft AI solution pattern-recognition
Trap: Goal-Driven Systems
Why wrong: Wrong-answer trap — Goal-Driven is optimization.
Correct answer was: C
Patterns & Anomalies
- D2-005 · Domain II · II.3 — Conduct risk assessment cross-domain-link
Trap: Just security risks
Why wrong: Wrong-answer trap — security alone.
Correct answer was: B
Cross-pull to Domain I: privacy risks (HIPAA), bias risks (health data has known bias patterns), regulatory risks (compliance with healthcare regulations), accountability risks (named human accountability for health decisions); ensure assessment is multi-dimensional
- D2-006 · Domain II · II.2 — Evaluate initial AI feasibility documentation-required
Trap: Required tech & skills
Why wrong: Wrong-answer trap — that's Technology Light 1.
Correct answer was: C
Sufficient data quality
- D2-007 · Domain II · II.6 — Manage adoption/integration risks iteration-trigger
Trap: Skip adoption planning since the AI is good
Why wrong: Wrong-answer trap — quality doesn't auto-drive adoption.
Correct answer was: B
Treat as adoption risk (II.6); engage stakeholders to develop adoption strategy: user training, change management, disclosure (Domain I.2), feedback mechanism, contestability path; document risk mitigation plan
- D2-008 · Domain II · II.4 — Develop AI project scope statement documentation-required
Trap: Just the budget
Why wrong: Wrong-answer trap — budget is one element.
Correct answer was: B
Problem statement, AI pattern (II.7), success criteria reference (II.8), in-scope/out-of-scope, key resources (II.10), key risks (II.3), trustworthy-AI constraints, assumptions, dependencies
- D2-009 · Domain II · II.10 — Identify project resources cross-domain-link
Trap: Just Domain IV
Why wrong: Wrong-answer trap — partial.
Correct answer was: B
Multiple: III.4 (workspace/infrastructure resources), V.1 (deployment plan resources), and IV.3 (training compute) all depend on II.10's resource identification
- D2-010 · Domain II · II.8 — Define success criteria go-no-go-gate
Trap: Have the data scientist adjust the model to meet original criteria
Why wrong: Wrong-answer trap — model adjustment without criterion review.
Correct answer was: B
Engage stakeholders to revise II.8 success criteria based on what's realistically achievable; document the revision and rationale; revised criteria flow into IV.6 gate evaluation; do not revise unilaterally
- D2-011 · Domain II · II.1 — Identify problem to be solved documentation-required
Trap: True — but only for B2C projects
Why wrong: Wrong-answer trap — persona applies broadly.
Correct answer was: A
True
- D2-012 · Domain II · II.2 — Evaluate initial AI feasibility pattern-recognition
Trap: Yellow follow-up needed
Why wrong: Wrong-answer trap — yellow logic.
Correct answer was: B
Stop. Do not proceed until resolved.
- D2-013 · Domain II · II.7 — Draft AI solution iteration-trigger
Trap: Have the data scientist change the pattern to Predictive Analytics
Why wrong: Wrong-answer trap — unilateral pattern change.
Correct answer was: B
Convene stakeholder review to revisit the pattern selection (II.7); pattern affects every downstream domain (III.1 data, IV.1 technique); document the revision and assess impact on II.4 scope
- D2-014 · Domain II · II.9 — Support business case creation documentation-required
Trap: PM signs the final case unilaterally
Why wrong: Wrong-answer trap — unilateral signing.
Correct answer was: B
PM coordinates the business case effort: gathering inputs from finance (ROI), Domain I (Trustworthy AI cost), data science (technical feasibility), business unit (value proposition); PM ensures comprehensive case with stakeholder engagement
- D2-015 · Domain II · II.8 — Define success criteria cross-domain-link
Trap: Just V.4 (production monitoring)
Why wrong: Wrong-answer trap — single point.
Correct answer was: B
Multiple: III.1 (data requirements include demographic representation), III.7 (data evaluation includes bias measurement), IV.2 (QA/QC includes bias), IV.6 (gate evaluates fairness), V.4 (production fairness monitoring), V.7 (contingency for fairness incidents) — fairness criterion threads through every weak domain
- D2-016 · Domain II · II.5 — Determine ROI documentation-required
Trap: False — AI ROI is always lower
Why wrong: Wrong-answer trap — overgeneralization.
Correct answer was: B
False — AI ROI often comes from probabilistic-decision improvement (error reduction value, decision-speed value, scalability value), not deterministic cost savings; AI ROI must include Trustworthy AI cost (audit, bias monitoring, governance, contestability)
- D2-018 · Domain II · II.10 — Identify project resources documentation-required
Trap: Just timeline
Why wrong: Wrong-answer trap — timeline is downstream.
Correct answer was: B
People, hardware, contractors — comprehensive resource identification
- D2-019 · Domain II · II.6 — Manage adoption/integration risks cross-domain-link
Trap: Skip opt-out since AI is good
Why wrong: Wrong-answer trap — quality doesn't waive consent.
Correct answer was: B
Cross-pull to Domain I (I.2 disclosure/consent): ensure consent mechanism with feasible alternatives is documented in II.6 plan and built into deployment (V.1) and contestability mechanism (I.4 + V.3)
- D2-020 · Domain II · II.7 — Draft AI solution pattern-recognition
Trap: Goal-Driven Systems
Why wrong: Wrong-answer trap — Goal-Driven is optimization.
Correct answer was: B
Conversational & Human Interaction
- D2-021 · Domain II · II.8 — Define success criteria documentation-required
Trap: True — but only quantitative
Why wrong: Wrong-answer trap — measurable includes both quantitative and structured qualitative.
Correct answer was: A
True
- D2-022 · Domain II · II.4 — Develop AI project scope statement iteration-trigger
Trap: Have the data scientist adjust
Why wrong: Wrong-answer trap — single role can't fix scope.
Correct answer was: B
Engage stakeholders to revise II.4 scope: descope features, extend timeline, add resources, or split into phases; document the revision; the choice is stakeholder decision, not unilateral PM call
- D2-023 · Domain II · II.5 — Determine ROI iteration-trigger
Trap: Hide the cost increase
Why wrong: Wrong-answer trap — hiding is integrity failure.
Correct answer was: B
Update ROI calculation to reflect actual data prep cost; engage stakeholders to revisit business case (II.9); if updated ROI no longer justifies the project, surface for go/no-go decision; document the revision
- D2-024 · Domain II · II.1 — Identify problem to be solved iteration-trigger
Trap: Have the data scientist build a hybrid system
Why wrong: Wrong-answer trap — hybrid without analysis.
Correct answer was: B
Loop back to Phase I (II.1) to revisit problem definition; engage stakeholders to assess: is rule-based automation more appropriate, is the AI scope still justified, can the project pivot? Document the decision; do not continue with mismatched solution.
- D2-025 · Domain II · II.7 — Draft AI solution documentation-required
Trap: Just the data sources
Why wrong: Wrong-answer trap — sources are partial.
Correct answer was: B
AI pattern, persona, key data needs (informs III.1), key technique direction (informs IV.1), trustworthy-AI considerations (Domain I), success criteria reference (II.8), risks identified (II.3) — the draft is comprehensive though high-level
- D2-026 · Domain II · II.3 — Conduct risk assessment cross-domain-link
Trap: Just workforce disruption
Why wrong: Wrong-answer trap — workforce alone.
Correct answer was: B
Ethics risks span: bias and fairness (I.3), human accountability (I.5), workforce disruption (I — Trustworthy AI Framework), societal/environmental impact (Trustworthy AI Framework), consent and disclosure (I.2); coordinate with ethics/legal/Domain I stakeholders for assessment
- D2-027 · Domain II · II.10 — Identify project resources iteration-trigger
Trap: Defer to operations
Why wrong: Wrong-answer trap — operations doesn't own pre-deployment resources.
Correct answer was: B
Coordinate with cloud/platform team to identify alternatives (different region, different instance type, different cloud); engage stakeholders if alternatives have cost/timeline impact; update II.10 resource plan; do not let unilateral resource change happen
- D2-028 · Domain II · II.6 — Manage adoption/integration risks documentation-required
Trap: True — marketing handles users
Why wrong: Wrong-answer trap — narrow framing.
Correct answer was: B
False — adoption risk is project-level (II.6) covering user training, change management, disclosure, feedback mechanisms, contestability; PM coordinates across functions including marketing
- D2-029 · Domain II · II.8 — Define success criteria go-no-go-gate
Trap: GO and address criteria concerns later
Why wrong: Wrong-answer trap — defer criteria revision.
Correct answer was: B
ITERATE back to II.8 to revise success criteria with stakeholders, then re-run III.8 gate; new criteria flow through every downstream evaluation; document both iterations
- D2-030 · Domain II · II.5 — Determine ROI documentation-required
Trap: Skip Domain II entirely
Why wrong: Wrong-answer trap — Domain II is 26% of exam.
Correct answer was: B
Refresh-only with PM Oversight Angle awareness applied retroactively, ROI/Business Case framing additions, and recognition that II.8 (Success Criteria) is the most cross-pulled artifact in the exam
- D3-001 · Domain III · III.1 — Define required data oversight-vs-doing
Trap: Direct the data scientist to start with the most accessible internal data and adjust later
Why wrong: Wrong-answer trap — convenience-based source selection bypasses requirements definition.
Correct answer was: C
Pause and produce a documented data requirements specification tying Phase I success criteria to specific data attributes, sources, and aggregation method, then engage stakeholders to confirm
- D3-002 · Domain III · III.8 — Determine if data meets solution needs go-no-go-gate
Trap: Have the data scientist engineer proxy features for the missing 20%
Why wrong: Wrong-answer trap — engineering proxy features bypasses governance.
Correct answer was: B
Conduct a formal go/no-go assessment, document the gap, evaluate impact on success criteria, and make a documented decision with stakeholders on whether to proceed-with-mitigation, iterate, or descope
- D3-003 · Domain III · III.6 — Check data privacy, compliance, and access cross-domain-link
Trap: Replace the partner data with a non-regulated substitute the data scientist suggests
Why wrong: Wrong-answer trap — substitute selection should be requirements-driven, not data-scientist-suggested.
Correct answer was: C
Document the delay as a risk, escalate to leadership with options (proceed-and-wait, iterate to alternative sources, or descope), and engage stakeholders for the decision
- D3-006 · Domain III · III.7 — Oversee data evaluation documentation-required
Trap: Skip the evaluation report and rely on training results to validate data fitness
Why wrong: Wrong-answer trap — deferring evaluation to training.
Correct answer was: B
Confirm a documented Data Evaluation Report exists covering completeness, quality across the seven dimensions, alignment with operational data, fitness for the AI pattern, and any remaining gaps; only then proceed to the gate
- D3-007 · Domain III · III.4 — Coordinate AI workspace and infrastructure escalation-trigger
Trap: Switch to on-premise infrastructure unilaterally
Why wrong: Wrong-answer trap — unilateral architecture change.
Correct answer was: B
Escalate the infrastructure capacity gap with cost/risk options to leadership; surface options including increased allocation, scope reduction, or timeline extension; engage stakeholders for the decision
- D3-008 · Domain III · III.8 — Determine if data meets solution needs go-no-go-gate
Trap: Have the data scientist apply data augmentation to make training data resemble operational data
Why wrong: Wrong-answer trap — augmentation is technical workaround.
Correct answer was: B
Recognize the training-vs-operational data mismatch as PMI iteration trigger #6; convene the III.8 gate with stakeholders; recommend ITERATE back to Phase I or III.1 to either re-source data, re-define the problem, or descope to a use case the available data supports
- D3-009 · Domain III · III.9 — Convey data understanding to leadership documentation-required
Trap: Send the data evaluation report to the data engineer and ML team to begin preparation
Why wrong: Wrong-answer trap — internal report bypasses leadership briefing.
Correct answer was: C
Convey Phase II findings (data understanding state, gate decision, key risks, recommendations, scope/schedule impacts) to leadership before Phase III work begins; ensure stakeholder authorization
- D3-010 · Domain III · III.6 — Check data privacy, compliance, and access cross-domain-link
Trap: Have the data engineer apply pseudonymization and proceed
Why wrong: Wrong-answer trap — pseudonymization is reversible, doesn't satisfy GDPR transfer.
Correct answer was: B
Coordinate with privacy/legal/compliance to assess GDPR cross-border transfer requirements (Article 44+); document the assessment; engage stakeholders if anonymization alone is insufficient or if alternative deployment locations are needed; do not proceed until the privacy plan is updated
- D3-012 · Domain III · III.3 — Identify data sources and locations documentation-required
Trap: Wait until renewal is decided before completing the source inventory
Why wrong: Wrong-answer trap — blocks inventory progress.
Correct answer was: B
Document the source AND escalate the contract risk for stakeholder visibility, since data availability is a project-level dependency that affects scope, schedule, and continuity
- D3-013 · Domain III · III.7 — Oversee data evaluation cross-domain-link
Trap: Have the data scientist add a fairness post-processing layer at deployment
Why wrong: Wrong-answer trap — post-processing is one option, not stakeholder substitute.
Correct answer was: B
Treat as ECO III.7 + I.3 (bias checks) + II.8 (success criteria) cross-domain issue: document the bias finding, engage stakeholders, evaluate mitigation options (re-source data with better representation, re-balance training data, in-processing fairness constraints, post-processing calibration), make a documented decision before proceeding
- D3-014 · Domain III · III.8 — Determine if data meets solution needs pattern-recognition
Trap: GO / NO-GO / On-Hold
Why wrong: Wrong-answer trap — On-Hold isn't an outcome.
Correct answer was: C
GO (proceed to Phase III) / ITERATE (loop back to address gaps) / DESCOPE (reduce scope to what data supports)
- D3-015 · Domain III · III.4 — Coordinate AI workspace and infrastructure oversight-vs-doing
Trap: Have the engineer use the project's existing on-premise infrastructure to avoid the security review
Why wrong: Wrong-answer trap — 'avoid security review' is governance bypass.
Correct answer was: B
Coordinate with platform/security/finance teams: confirm capacity request aligns with the documented infrastructure plan; ensure security review is engaged for the new cloud region; surface cost decision to stakeholders for authorization
- D3-017 · Domain III · III.5 — Gather required data documentation-required
Trap: Have the data scientist begin evaluation immediately to save time
Why wrong: Wrong-answer trap — parallel evaluation without verification.
Correct answer was: B
Confirm the Data Source Inventory has been updated with status per source (gathered / partial / blocked); confirm the staged dataset matches III.1 requirements; document any gathering blockers and their resolution
- D3-018 · Domain III · III.1 — Define required data iteration-trigger
Trap: Approve the pattern change since the data scientist is closest to the technical decision
Why wrong: Wrong-answer trap — unilateral pattern change.
Correct answer was: B
Loop back to Phase I (Business Understanding / Domain II) to reconsider the AI pattern selection (II.7) with stakeholders, since the pattern is a Phase I deliverable and changing it has scope implications; do not redefine the pattern unilaterally inside Domain III
- D3-019 · Domain III · III.8 — Determine if data meets solution needs go-no-go-gate
Trap: Conditional GO with a note that privacy plan will be signed off in Phase III
Why wrong: Wrong-answer trap — conditional GOs become governance debt.
Correct answer was: B
Recognize that III.8 includes trustworthy-AI constraints (cross-pull to I.1 privacy plan); document the privacy plan gap; ITERATE — do not GO until privacy plan sign-off is complete
- D3-021 · Domain III · III.6 — Check data privacy, compliance, and access escalation-trigger
Trap: Have the data engineer apply pseudonymization — a stronger anonymization technique — and proceed
Why wrong: Wrong-answer trap — pseudonymization (reversible) doesn't address consent.
Correct answer was: B
Treat as a serious compliance and accountability incident; escalate to legal/compliance/leadership; document the consent gap; consider the data tainted for this project until legal review concludes whether the data can be used (with or without anonymization), whether consent can be obtained retroactively, or whether the data must be excluded; do not proceed unilaterally
- D3-022 · Domain III · III.2 — Identify data SMEs pattern-recognition
Trap: Stewards and custodians are the same role with different titles
Why wrong: Wrong-answer trap — conflating roles is common misconception.
Correct answer was: C
Stewards enforce policy and curate data (strategic, policy-enforcing, cross-functional); custodians ensure safe storage, transfer, and use of data (operational, administrative, NOT the data owner)
- D3-023 · Domain III · III.5 — Gather required data iteration-trigger
Trap: Have the data engineer try to gather more data from external sources unilaterally
Why wrong: Wrong-answer trap — unilateral external sourcing.
Correct answer was: B
Pause and recognize PMI iteration trigger #4/5 (data too little / augmentation considered). Engage stakeholders to evaluate options: synthetic data with QA, additional sourcing, technique change, or descope. Document the decision before proceeding.
- D3-024 · Domain III · III.4 — Coordinate AI workspace and infrastructure cross-domain-link
Trap: Have the ML engineer document the manual installation steps for repeatability
Why wrong: Wrong-answer trap — documentation doesn't fix brittleness.
Correct answer was: B
Recognize this as cross-domain pull (III.4 × IV.1 × V.1): coordinate with platform team to either provision a custom image or evaluate whether the technique should be reconsidered (loop back to IV.1); ensure governance and reproducibility (V.3) are not compromised
- D3-026 · Domain III · III.1 — Define required data documentation-required
Trap: Proceed without the linkages and note them in lessons learned
Why wrong: Wrong-answer trap — deferring linkage compounds problems.
Correct answer was: B
Revise the data requirements specification to explicitly tie each required field to the Phase I AI pattern and Domain II success criteria, then engage the data scientist and stakeholders to validate
- D3-027 · Domain III · III.3 — Identify data sources and locations oversight-vs-doing
Trap: Have the data scientist proceed and document the source after
Why wrong: Wrong-answer trap — documenting after is too late.
Correct answer was: B
Coordinate verification of license, provenance, bias measurement availability, and privacy/compliance fit before adding the dataset to the documented Data Source Inventory
- D3-028 · Domain III · III.7 — Oversee data evaluation documentation-required
Trap: Have the data scientist define their own success criteria
Why wrong: Wrong-answer trap — data scientist defining own success criteria bypasses governance.
Correct answer was: B
Provide a documented evaluation rubric tied to Domain II success criteria covering completeness, quality dimensions, alignment with operational data, fitness for the AI pattern, bias measurements, and gap identification
- D3-029 · Domain III · III.8 — Determine if data meets solution needs go-no-go-gate
Trap: GO at the gate and let model performance reveal whether volume was sufficient
Why wrong: Wrong-answer trap — letting model performance reveal volume issue means wasted training cycles.
Correct answer was: B
ITERATE — engage stakeholders to evaluate options: pretrained model with fine-tuning (less data needed), data augmentation, alternative technique (IV.1 loop), or descope. Document the decision.
- D3-030 · Domain III · III.6 — Check data privacy, compliance, and access cross-domain-link
Trap: Skip the training and document the deviation
Why wrong: Wrong-answer trap — documented deviation is still governance bypass.
Correct answer was: B
Confirm the training is part of the data steward's documented governance program and incorporate it into the project timeline; do not skip; the request itself signals a governance discipline gap that needs to be reinforced
- D3-032 · Domain III · III.5 — Gather required data escalation-trigger
Trap: Have the data engineer redact identifying information and use the data
Why wrong: Wrong-answer trap — redaction doesn't override contract prohibition.
Correct answer was: B
Treat as a serious compliance and contractual incident; escalate to legal/compliance/contract-management; document the prohibition; remove the data from the project unless legal review concludes alternative permitted uses; consider this a III.6 violation
- D3-033 · Domain III · III.2 — Identify data SMEs oversight-vs-doing
Trap: Have the data scientist email the SME and proceed
Why wrong: Wrong-answer trap — uncoordinated communication.
Correct answer was: B
Coordinate scheduling, async communication, or alternative SMEs; document SME engagement plan with cadence; do not let logistical friction block engagement on a project-critical dependency
- D3-035 · Domain III · III.7 — Oversee data evaluation go-no-go-gate
Trap: Side with the data scientist since they know the data best
Why wrong: Wrong-answer trap — siding with one stakeholder isn't facilitation.
Correct answer was: B
Facilitate documented analysis of each stakeholder's concern with criteria-based evaluation; surface the trade-offs to leadership; recommend ITERATE if any criterion is materially short, since the cost of unresolved gaps compounds in Phase III/IV/V
- D3-036 · Domain III · III.6 — Check data privacy, compliance, and access documentation-required
Trap: Have the data engineer anonymize data so disclosure isn't needed
Why wrong: Wrong-answer trap — anonymization doesn't always exempt disclosure.
Correct answer was: B
Pause data prep until the privacy policy update is complete; the disclosure requirement is part of the privacy plan (I.1) and proceeding before disclosure is policy violation; surface schedule impact to stakeholders
- D3-037 · Domain III · III.5 — Gather required data iteration-trigger
Trap: Approve imputation since it's a standard technique
Why wrong: Wrong-answer trap — approving without quality/bias evaluation.
Correct answer was: B
Document the missingness, escalate to III.7 evaluation to assess whether imputation will preserve data quality and bias tolerance, engage stakeholders for the decision (proceed-with-imputation, re-source, or descope) before authorizing imputation
- D3-038 · Domain III · III.3 — Identify data sources and locations cross-domain-link
Trap: Have the data scientist anonymize prompts and proceed
Why wrong: Wrong-answer trap — prompts may contain PII even when fields are anonymized.
Correct answer was: B
Recognize this as a Domain I cross-pull (privacy I.1, compliance I.4, transparency I.2): assess whether sending customer prompts violates privacy plan, regulatory commitments, or disclosure obligations; engage privacy/legal; consider self-hosted alternatives if the public LLM doesn't meet requirements
- D3-039 · Domain III · III.4 — Coordinate AI workspace and infrastructure escalation-trigger
Trap: Find a workaround using only approved services
Why wrong: Wrong-answer trap — workaround may compromise the technical solution.
Correct answer was: B
Coordinate the security exception request with security/architecture/compliance; provide project context; surface alternative approaches if the exception is denied or delayed; document the exception process and outcome for governance traceability
- D3-040 · Domain III · III.8 — Determine if data meets solution needs iteration-trigger
Trap: Defer the success-criteria revision to Phase IV
Why wrong: Wrong-answer trap — deferring criteria revision compounds risk.
Correct answer was: B
Recommend ITERATE back to II.8 (revise success criteria) before GO; engage stakeholders to revise success criteria based on what the data realistically supports; then re-run the III.8 gate against revised criteria; document both rounds
- D3-041 · Domain III · III.1 — Define required data cross-domain-link
Trap: Note the fairness criterion separately for later
Why wrong: Wrong-answer trap — separation defers cross-pull.
Correct answer was: B
Required demographic representation in the data, bias-measurement requirements (cross-pull to I.3), and attribute-level detail that allows fairness verification — all tied to II.8 success criteria
- D3-044 · Domain III · III.2 — Identify data SMEs oversight-vs-doing
Trap: True — custodian is the legal owner
Why wrong: Wrong-answer trap — 'legal owner' framing.
Correct answer was: B
False — a data custodian serves in an administrative role over the data, NOT as the data owner. They handle safe storage, transfer, and use.
- D3-046 · Domain III · III.8 — Determine if data meets solution needs go-no-go-gate
Trap: Document the hallway conversation retroactively
Why wrong: Wrong-answer trap — retroactive documentation papers over the missing process.
Correct answer was: B
Convene a formal documented gate review with stakeholders; even if the decision is the same, the documentation is the artifact III.8 produces; verbal-only decisions don't satisfy the task
- D3-047 · Domain III · III.1 — Define required data documentation-required
Trap: Validation and test data are Phase IV concerns
Why wrong: Wrong-answer trap — split spec across phases creates gaps.
Correct answer was: B
All three (training, validation, test) plus production-inference data should be defined; the spec covers volumes, sources, and quality criteria for each — not just training
- D3-048 · Domain III · III.3 — Identify data sources and locations iteration-trigger
Trap: Have the data engineer estimate the impact unilaterally
Why wrong: Wrong-answer trap — engineer-only impact estimation lacks stakeholder context.
Correct answer was: B
Loop back to III.3 to evaluate additional sources (other internal systems, external feeds, partner data, public datasets, synthetic data); engage stakeholders if external acquisition has cost/contract implications; document the source-inventory revision
- D3-049 · Domain III · III.7 — Oversee data evaluation cross-domain-link
Trap: Have the data scientist build drift monitoring tools now
Why wrong: Wrong-answer trap — building tools now is premature; documenting expectations is the right step.
Correct answer was: B
Recognize that drift expectations should inform the deployment plan (V.1) and contingency plan (V.7) being designed in parallel; document drift expectations now; ensure they shape the V.1/V.4/V.7 work in Phase VI; do not 'defer' considerations that affect downstream design
- D3-050 · Domain III · III.4 — Coordinate AI workspace and infrastructure documentation-required
Trap: The data scientist's preferred tools
Why wrong: Wrong-answer trap — preferences without governance.
Correct answer was: B
Provisioning plan + status tracker covering compute, storage, pipelines, environments (dev/test/prod), security controls, access provisioning, and stakeholder sign-off from platform/security teams
- D3-052 · Domain III · III.9 — Convey data understanding to leadership escalation-trigger
Trap: Have the data scientist provide the timeline assessment
Why wrong: Wrong-answer trap — data scientist doesn't own project timeline.
Correct answer was: B
Honestly communicate the data state, including risks, and recommend alternatives if the Q3 timeline is at risk: descope, additional resources, or schedule revision; engage leadership for the decision
- D3-053 · Domain III · III.6 — Check data privacy, compliance, and access pattern-recognition
Trap: Pseudonymization is stronger than anonymization
Why wrong: Wrong-answer trap — common misconception. Anonymization is stronger (irreversible).
Correct answer was: B
Anonymization is irreversible (subjects cannot be re-identified); pseudonymization is reversible (subjects CAN be re-identified by someone with the key). GDPR treats them differently — pseudonymized data is still personal data.
- D3-054 · Domain III · III.1 — Define required data iteration-trigger
Trap: Have the data engineer build a real-time pipeline
Why wrong: Wrong-answer trap — unilateral building of real-time pipeline without scope/cost decision.
Correct answer was: B
Engage stakeholders: surface the gap between II.8's real-time requirement and the organization's current data-collection capability; options include investing in real-time infrastructure, revising II.8 to batch-acceptable criteria, or descoping; loop back to Phase I if business problem changes
- D3-055 · Domain III · III.3 — Identify data sources and locations documentation-required
Trap: Have the data engineer fill in their best guess
Why wrong: Wrong-answer trap — best-guess on access creates security/compliance gaps.
Correct answer was: B
Resolve the access mechanism for each entry before submitting; an inventory with undefined access is incomplete and creates III.5 (gathering) blockers
- D3-056 · Domain III · III.7 — Oversee data evaluation cross-domain-link
Trap: Trustworthy AI is Phase VI deployment concern
Why wrong: Wrong-answer trap — Phase VI is too late for trust evaluation.
Correct answer was: B
Trustworthy-AI considerations (bias measurement, privacy verification, transparency capability) ARE part of III.7 evaluation; cross-pulls from Domain I.1, I.2, I.3 belong in evaluation rubric; without them, III.8 gate is incomplete
- D3-057 · Domain III · III.2 — Identify data SMEs documentation-required
Trap: Email addresses only
Why wrong: Wrong-answer trap — contact info isn't engagement plan.
Correct answer was: B
Required data element × SME role × name × engagement scope (which Domain III tasks they support — requirements review, source validation, evaluation, gate input) × cadence × backup SME
- D3-058 · Domain III · III.8 — Determine if data meets solution needs go-no-go-gate
Trap: Have the data scientist proceed as pilot but flag the issue in lessons learned
Why wrong: Wrong-answer trap — flag-and-proceed isn't the iteration response.
Correct answer was: B
Recognize PMI iteration trigger #12 (PoC vs Pilot misalignment); loop back to Phase I to redefine project scope as PoC with explicit follow-up plan, OR descope to confirm pilot data requirements; do not proceed as pilot with PoC-only data
- D3-059 · Domain III · III.6 — Check data privacy, compliance, and access iteration-trigger
Trap: Have the data engineer restrict access logs to mask the violation
Why wrong: Wrong-answer trap — masking is governance violation worse than the original.
Correct answer was: B
Loop back to III.1 (re-define required data) or escalate for approval extension; do not use restricted data outside approved use cases without documented authorization; engage data steward and policy owners for the decision
- D3-060 · Domain III · III.4 — Coordinate AI workspace and infrastructure phase-identification
Trap: Infrastructure isn't a Domain III concern
Why wrong: Wrong-answer trap — III.4 explicitly covers infrastructure.
Correct answer was: B
Workspace + dev/test environments + data pipelines are III.4. Production deployment infrastructure choices are V.1 — but operational fit must be verified at IV.6 gate. III.4 decisions inform but don't substitute for V.1.
- D3-061 · Domain III · III.1 — Define required data go-no-go-gate
Trap: There's no formal criterion
Why wrong: Wrong-answer trap — informal criteria create governance gaps.
Correct answer was: B
The spec is sufficient when it covers AI pattern reference, required data types, attribute-level detail, source list, aggregation plan, and trustworthy-AI constraints — and is signed off by data SMEs, the data scientist, and project stakeholders. Insufficient = ITERATE.
- D3-062 · Domain III · III.5 — Gather required data cross-domain-link
Trap: Use all the data and document consent variability
Why wrong: Wrong-answer trap — using non-consenting data is privacy violation.
Correct answer was: B
Coordinate with privacy/legal/compliance: assess consent-tier differences against project use; document consent strata; engage stakeholders for the decision (use consenting-only, expand consent, or descope); update privacy plan; do not gather across consent tiers without legal clearance
- D3-063 · Domain III · III.8 — Determine if data meets solution needs documentation-required
Trap: Just the gate decision (GO)
Why wrong: Wrong-answer trap — decision-only without rationale isn't governance.
Correct answer was: B
A Phase II Go/No-Go Decision Document covering: source findings, description findings (4 Vs), quality findings, gap analysis, decision (GO), rationale, stakeholders engaged, sign-off, downstream actions for Phase III initiation
- D3-064 · Domain III · III.7 — Oversee data evaluation oversight-vs-doing
Trap: Have the data scientist install and use the tool informally
Why wrong: Wrong-answer trap — informal use is license violation.
Correct answer was: B
Coordinate license review with legal/procurement; evaluate alternatives (open-source tools with commercial-friendly licenses); ensure license terms align with project use; do not install commercial-restricted tools without authorization
- D3-066 · Domain III · III.6 — Check data privacy, compliance, and access documentation-required
Trap: Just the PM
Why wrong: Wrong-answer trap — PM coordinates but doesn't unilaterally sign off.
Correct answer was: B
Privacy officer, legal counsel, compliance team, data steward (where applicable), and project sponsor — depending on data sensitivity and regulatory framework. PM coordinates the sign-off.
- D3-067 · Domain III · III.2 — Identify data SMEs iteration-trigger
Trap: Note it in lessons learned and proceed
Why wrong: Wrong-answer trap — flag-and-proceed loses the SME-gap implications.
Correct answer was: B
Surface the SME gap as a project risk; engage stakeholders for options: identify a substitute SME from related domains, hire an external SME, descope the dataset's use, or accept a documented knowledge-gap risk; document the decision
- D3-069 · Domain III · III.4 — Coordinate AI workspace and infrastructure iteration-trigger
Trap: Bypass the delayed vendor and use an unauthorized alternative
Why wrong: Wrong-answer trap — unauthorized alternatives are governance violations.
Correct answer was: B
Surface the delay as a project risk; evaluate alternatives (different vendor, internal capacity, scope reduction, schedule revision); engage stakeholders for the decision; document the impact and the chosen path
- D3-070 · Domain III · III.1 — Define required data oversight-vs-doing
Trap: True — once technical needs are listed
Why wrong: Wrong-answer trap — technical list isn't governance-complete spec.
Correct answer was: B
False — III.1 is complete when there's a documented requirements specification tying Phase I success criteria to specific data attributes, sources, and aggregation method, with stakeholder sign-off (data SMEs, data scientist, project stakeholders)
- D3-071 · Domain III · III.5 — Gather required data documentation-required
Trap: Once a week
Why wrong: Wrong-answer trap — weekly is too coarse for active gathering.
Correct answer was: B
Continuously as gathering progresses — status per source (gathered, partial, blocked) updated in real time so the PM can surface blockers and stakeholders have visibility
- D3-072 · Domain III · III.7 — Oversee data evaluation documentation-required
Trap: Skip the bias section and add it later
Why wrong: Wrong-answer trap — 'add later' loses the gate input.
Correct answer was: B
Insist that bias measurement is part of III.7 evaluation (cross-pull to I.3); coordinate with the team to either get tooling ready, use an alternative measurement approach, or pause evaluation until tooling exists; do not produce an evaluation report missing required sections
- D3-073 · Domain III · III.6 — Check data privacy, compliance, and access go-no-go-gate
Trap: GO with conditional sign-off pending legal review
Why wrong: Wrong-answer trap — conditional GO with pending legal is governance debt.
Correct answer was: B
ITERATE — recognize that III.6 is a continuous-monitoring task (cross-pull to I.4); legal's regulatory assessment is in scope; pause GO until legal review concludes or risk-tolerance is documented at leadership level
- D3-074 · Domain III · III.2 — Identify data SMEs cross-domain-link
Trap: Have the data scientist serve as both
Why wrong: Wrong-answer trap — single person serving both is too narrow.
Correct answer was: B
Engage BOTH technical and business SMEs; technical SMEs validate data structure/access; business SMEs validate that the data captures the business meaning correctly; both perspectives feed III.7 evaluation and III.8 gate
- D3-075 · Domain III · III.8 — Determine if data meets solution needs iteration-trigger
Trap: GO and address the new risks in Phase III
Why wrong: Wrong-answer trap — Phase III isn't risk-reassessment venue.
Correct answer was: B
Pause and convene a documented risk re-assessment with stakeholders; the new external information may invalidate Phase I risk assumptions; consider ITERATE back to Phase I to re-assess (II.3 risk assessment); leadership decision required
- D4-001 · Domain IV · IV.5 — Verify data quality (GATE) go-no-go-gate
Trap: Have the data scientist start training in parallel with the gate review
Why wrong: Wrong-answer trap — parallel work bypasses gate.
Correct answer was: B
Run the IV.5 verification gate — quality dimensions, coverage, bias, volume, reproducibility — with stakeholders before authorizing training
- D4-002 · Domain IV · IV.3 — Manage AI/ML model training iteration-trigger
Trap: Have them switch to a smaller model immediately
Why wrong: Wrong-answer trap — switching technique without IV.1 review.
Correct answer was: B
Pause training, conduct structured review of root cause (data, technique, hardware, results), reassess against project plan, document decision on whether to continue, change approach, or escalate
- D4-003 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) go-no-go-gate
Trap: Have the ML engineer start deployment while the gate is being scheduled
Why wrong: Wrong-answer trap — parallel work bypasses gate.
Correct answer was: B
Run the IV.6 operationalization-readiness gate with stakeholders, evaluating performance, bias, robustness, baseline, audit, reproducibility, trustworthy-AI alignment, and operational fit
- D4-005 · Domain IV · IV.1 — Oversee AI/ML model technique(s) documentation-required
Trap: Have the data scientist proceed and add post-hoc XAI explanations after training
Why wrong: Wrong-answer trap — post-hoc XAI may not satisfy regulator's bar.
Correct answer was: B
Document the technique selection and ensure trade-off between performance and explainability is presented to stakeholders for decision; consider interpretable-by-design alternatives
- D4-006 · Domain IV · IV.2 — Oversee AI/ML model QA/QC cross-domain-link
Trap: Have the data scientist add a fairness post-processing layer
Why wrong: Wrong-answer trap — post-processing without governance.
Correct answer was: B
Treat as IV.2 + I.3 (bias checks) issue: document, escalate per accountability procedures, engage stakeholders for remediation decision, do not authorize IV.6 GO until bias is within tolerance
- D4-007 · Domain IV · IV.3 — Manage AI/ML model training iteration-trigger
Trap: Have the data scientist try a different algorithm
Why wrong: Wrong-answer trap — algorithm change without root-cause.
Correct answer was: B
Pause and review the iteration trajectory: are improvements converging or plateauing? Is the technique a fit? Is the data sufficient? Document a decision: continue, change technique, descope, or escalate.
- D4-008 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) go-no-go-gate
Trap: Have the ML engineer optimize the model for CPU
Why wrong: Wrong-answer trap — workaround without IV.1 documentation.
Correct answer was: B
Treat as operational-fit failure of IV.6 — documented ITERATE outcome. Loop back to V.1 to address infrastructure OR loop back to IV.1 to choose a model that fits the planned environment.
- D4-010 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) go-no-go-gate
Trap: Approve parallel work to save time
Why wrong: Wrong-answer trap — 'save time' rationalization.
Correct answer was: B
Confirm IV.6 must complete before Domain V work begins; sequential, not parallel; gate authorizes the transition
- D4-011 · Domain IV · IV.1 — Oversee AI/ML model technique(s) documentation-required
Trap: True — automation = governance compliance
Why wrong: Wrong-answer trap — automation ≠ documentation.
Correct answer was: B
False — AutoML automates technical selection but doesn't replace governance. PM still needs the chosen technique documented, justified against AI pattern + success criteria, aligned with operational constraints. AutoML output feeds IV.1 documentation; doesn't bypass.
- D4-012 · Domain IV · IV.5 — Verify data quality (GATE) go-no-go-gate
Trap: Conditional GO with bias remediation in training
Why wrong: Wrong-answer trap — conditional GO with bias is governance debt.
Correct answer was: B
ITERATE — bias is part of criteria. Bias outside tolerance means IV.5 doesn't GO. Loop to IV.4 to remediate or III.1 to redefine requirements.
- D4-013 · Domain IV · IV.2 — Oversee AI/ML model QA/QC documentation-required
Trap: Just final evaluation
Why wrong: Wrong-answer trap — final-only misses continuous discipline.
Correct answer was: B
Configuration management (versions of code, data, model, hyperparameters, environment) + performance verification + bias measurement + documentation throughout development
- D4-014 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) documentation-required
Trap: Performance, Build quality, Resilience, Bias, Audit, Reliability, Test coverage, Output quality
Why wrong: Wrong-answer trap — different mapping.
Correct answer was: B
Performance, Bias, Robustness, Baseline, Audit, Reproducibility, Trustworthy AI, Operational fit
- D4-015 · Domain IV · IV.3 — Manage AI/ML model training documentation-required
Trap: Security team
Why wrong: Wrong-answer trap — security relevant for production but not training execution.
Correct answer was: B
Data scientist (ML engineer assists with infrastructure); PM manages — coordinates, tracks, surfaces blockers
- D4-016 · Domain IV · IV.5 — Verify data quality (GATE) iteration-trigger
Trap: Have the engineer document the specific machine setup
Why wrong: Wrong-answer trap — documentation doesn't equal production-grade.
Correct answer was: B
ITERATE — reproducibility per means production-grade, not single-machine. Loop back to IV.4 to validate on production-like infrastructure; loop to III.4 if infrastructure changes are needed; do not GO.
- D4-017 · Domain IV · IV.1 — Oversee AI/ML model technique(s) cross-domain-link
Trap: Just the data scientist's preference
Why wrong: Wrong-answer trap — preference isn't governance.
Correct answer was: B
Justification: AI pattern fit (II.7), success criteria fit (II.8), trustworthy-AI implications (Domain I — privacy, transparency, bias, governance), operational fit (V.1), cost, and explainability — engaging multi-stakeholder review
- D4-018 · Domain IV · IV.2 — Oversee AI/ML model QA/QC cross-domain-link
Trap: Note the gap in lessons learned
Why wrong: Wrong-answer trap — note without remediation.
Correct answer was: B
Recognize as Domain I.5 (accountability/audit) cross-pull; coordinate retroactive audit-trail reconstruction where feasible; document the gap and remediation; the missing training-audit may invalidate IV.6 reproducibility check; engage stakeholders before IV.6 GO
- D4-019 · Domain IV · IV.4 — Manage data transformation to conduct data preparation iteration-trigger
Trap: Have the engineer fix the rule and continue
Why wrong: Wrong-answer trap — fix-and-continue without root-cause.
Correct answer was: B
Coordinate root-cause investigation (is it data error, rule error, edge cases?); engage data scientist + SMEs to evaluate impact on training; engage stakeholders for the decision (fix data, fix rule, exclude records, or descope); do not unilaterally drop 10% of training data
- D4-020 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) cross-domain-link
Trap: Note the deviation and GO
Why wrong: Wrong-answer trap — GO with deviation.
Correct answer was: B
ITERATE — unauthorized data is a III.1 spec deviation and a governance issue (I.5 accountability). Document the deviation, evaluate whether the unauthorized data introduces compliance/privacy/bias risks, engage stakeholders, retrain on spec-compliant data if needed
- D4-022 · Domain IV · IV.5 — Verify data quality (GATE) go-no-go-gate
Trap: Defer all V.1 work until IV.6 GO
Why wrong: Wrong-answer trap — V.1 drafting can begin earlier; finalization at IV.6.
Correct answer was: B
Note that V.1 deployment plan can begin drafting earlier (it's iterative through the project), but Domain V execution waits for IV.6 GO; IV.5 GO authorizes IV.3 training work, not Domain V work
- D4-023 · Domain IV · IV.3 — Manage AI/ML model training documentation-required
Trap: Just the time spent
Why wrong: Wrong-answer trap — time alone insufficient.
Correct answer was: B
Training Status Tracker covering progress, hyperparameter choices, training data version, environment configuration, intermediate metrics, blockers and resolutions; supports IV.2 QA/QC and IV.6 reproducibility
- D4-024 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) documentation-required
Trap: Adjust the success criteria
Why wrong: Wrong-answer trap — unilateral criterion adjustment.
Correct answer was: B
ITERATE — performance vs baseline is part of. 2 vs 5 expected = performance below criteria. Engage stakeholders to revise success criteria (Domain II) or improve the model (loop IV.3) or descope.
- D4-025 · Domain IV · IV.2 — Oversee AI/ML model QA/QC iteration-trigger
Trap: Have the data scientist generate synthetic OOD data
Why wrong: Wrong-answer trap — synthetic OOD is approximation.
Correct answer was: B
Coordinate evaluation against an out-of-distribution (OOD) test set — production data may differ from training distribution; engage data scientist to create OOD evaluation; document findings before IV.6 GO
- D4-026 · Domain IV · IV.4 — Manage data transformation to conduct data preparation documentation-required
Trap: False — IV.4 is never complete
Why wrong: Wrong-answer trap — flippant.
Correct answer was: B
False — IV.4 completion = documented Data Preparation Plan executed + status tracker complete + prepared dataset usable + IV.5 gate confirmation
- D4-027 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) iteration-trigger
Trap: Have the data scientist write up the gate as a formality
Why wrong: Wrong-answer trap — formality without stakeholder engagement.
Correct answer was: B
Convene the documented gate review even though the team is confident; the gate is the methodology's stakeholder-engagement and accountability artifact, not a quality check on the team's judgment; produce documented decision regardless of foregone conclusion
- D4-028 · Domain IV · IV.1 — Oversee AI/ML model technique(s) iteration-trigger
Trap: Switch to open-source and proceed
Why wrong: Wrong-answer trap — open-source still has license terms; doesn't auto-resolve patent.
Correct answer was: B
Coordinate with legal/IP team to assess patent applicability; document findings; engage stakeholders for the decision (proceed-with-license, alternative architecture, descope, or accept documented risk); do not approve technique with unresolved IP issue
- D4-029 · Domain IV · IV.5 — Verify data quality (GATE) documentation-required
Trap: An email to the team
Why wrong: Wrong-answer trap — informal communication.
Correct answer was: B
Phase III Go/No-Go Decision Document with quality findings, coverage, bias measurements, volume assessment, reproducibility verification, decision (ITERATE), specific gaps to address, target loop (IV.4 / III.7 / III.1 / Phase I), stakeholder sign-off, and downstream actions
- D4-030 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) go-no-go-gate
Trap: Conditional GO with regulatory review pending
Why wrong: Wrong-answer trap — conditional GO with pending legal.
Correct answer was: B
ITERATE pending legal assessment; regulatory shift affects Trustworthy AI alignment ( criterion); pause GO until legal review concludes or risk-tolerance is documented at leadership level
- D4-032 · Domain IV · IV.3 — Manage AI/ML model training cross-domain-link
Trap: Have the data scientist use the LLM informally
Why wrong: Wrong-answer trap — informal use is governance bypass.
Correct answer was: B
Coordinate Domain I cross-pull: privacy plan (I.1) — does sending data to public LLM violate plan? Compliance (I.4) — regulatory exposure? Engage privacy/legal/security; consider self-hosted alternatives if public LLM doesn't meet requirements
- D4-033 · Domain IV · IV.5 — Verify data quality (GATE) iteration-trigger
Trap: Have the data engineer gather more data unilaterally
Why wrong: Wrong-answer trap — unilateral data acquisition without scope decision.
Correct answer was: B
ITERATE on technique (loop IV.1) since volume is insufficient for chosen technique; alternative: GO if technique is changed to a model that fits data volume; document the constraint and decision
- D4-034 · Domain IV · IV.2 — Oversee AI/ML model QA/QC documentation-required
Trap: Only on production data
Why wrong: Wrong-answer trap — production-only misses development opportunity.
Correct answer was: B
Throughout development — during data prep evaluation (III.7), during QA/QC (IV.2 continuous), at the IV.6 gate, and into production monitoring (V.4); single-snapshot measurement is insufficient
- D4-035 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) documentation-required
Trap: Project closeout
Why wrong: Wrong-answer trap — closeout is way later.
Correct answer was: B
Domain V work to begin: V.1 deployment plan finalization, V.2 deployment execution, V.3 governance, etc. — IV.6 is the bridge to Domain V.
- D4-036 · Domain IV · IV.1 — Oversee AI/ML model technique(s) documentation-required
Trap: Just the data scientist's reasoning
Why wrong: Wrong-answer trap — single-perspective.
Correct answer was: B
Technique justification: algorithm/family/pretrained choices, rationale, alignment with Phase I AI pattern, alignment with Domain II success criteria, operational implications (cost, latency, explainability), trustworthy-AI implications, alternatives considered and rejected
- D4-037 · Domain IV · IV.5 — Verify data quality (GATE) cross-domain-link
Trap: Have the data scientist re-run bias measurement on test set
Why wrong: Wrong-answer trap — test set ≠ operational.
Correct answer was: B
Recognize as cross-domain pull (IV.5 + I.3 + III.7); the bias measurement should reflect operational distribution; ITERATE to perform operational-distribution-aligned bias measurement; without it, IV.5 GO is on incorrect baseline
- D4-038 · Domain IV · IV.3 — Manage AI/ML model training iteration-trigger
Trap: Have the data scientist try one more iteration
Why wrong: Wrong-answer trap — blind iteration.
Correct answer was: B
Coordinate with the data scientist to investigate early plateau (data quality, technique fit, hyperparameter tuning); engage IV.2 QA/QC; if early plateau is suspicious of underfitting or data limitation, surface to IV.6 review; do not blindly accept 'acceptable' performance without root-cause
- D4-039 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) iteration-trigger
Trap: Note the gap and GO
Why wrong: Wrong-answer trap — note-and-proceed.
Correct answer was: B
Recognize that adversarial robustness is part of Robustness criterion in; if not measured, robustness is incomplete; coordinate retroactive adversarial-robustness evaluation; consider as ITERATE or document risk acceptance; engage Domain I (security) and stakeholders
- D4-040 · Domain IV · IV.5 — Verify data quality (GATE) documentation-required
Trap: An email to the team
Why wrong: Wrong-answer trap — informal communication.
Correct answer was: B
Phase III Go/No-Go Decision Document findings, decision (GO), rationale, stakeholder sign-off, downstream action (IV.3 training authorized to begin)
- D4-042 · Domain IV · IV.4 — Manage data transformation to conduct data preparation documentation-required
Trap: Try, Review, Implement, Modify
Why wrong: Wrong-answer trap — generic mapping.
Correct answer was: B
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: Have the data scientist proceed and add baseline comparison later
Why wrong: Wrong-answer trap — defer baseline.
Correct answer was: B
ITERATE — Baseline comparison is part of; required regardless of AI performance. If AI doesn't beat the rule-based baseline meaningfully, the project ROI is questionable. Coordinate baseline implementation and comparison.
- D4-044 · Domain IV · IV.1 — Oversee AI/ML model technique(s) iteration-trigger
Trap: Have the data scientist use the model informally
Why wrong: Wrong-answer trap — informal use bypasses governance.
Correct answer was: B
Coordinate provenance review per Domain I principles: training data origin, license compatibility, bias measurement availability, governance support; engage legal/compliance/security; consider alternatives if provenance is insufficient; do not approve undocumented training data without governance review
- D4-045 · Domain IV · IV.3 — Manage AI/ML model training documentation-required
Trap: Document, Train, Hand-off, Refactor
Why wrong: Wrong-answer trap — different mapping.
Correct answer was: B
Data (quality, volume, distribution), Technique (algorithm fit), Hardware/resources, Results so far
- D4-046 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) cross-domain-link
Trap: GO without privacy sign-off
Why wrong: Wrong-answer trap — bypassing privacy review.
Correct answer was: B
Coordinate with privacy/legal team for an interim sign-off authority (delegate, deputy, alternative); document the delegation; engage with the original officer when they return; do not let single-person availability block IV.6 indefinitely while ensuring privacy review is still real
- D4-047 · Domain IV · IV.2 — Oversee AI/ML model QA/QC documentation-required
Trap: False — IV.2 is the operations team's job
Why wrong: Wrong-answer trap — operations is post-deployment.
Correct answer was: B
False — IV.2 is PM-overseen, multi-team: data scientist runs technical QA, ML engineer handles config management, data steward validates data quality, security validates security QA, all coordinated by PM
- D4-049 · Domain IV · IV.3 — Manage AI/ML model training documentation-required
Trap: When the data scientist says it's done
Why wrong: Wrong-answer trap — single-perspective.
Correct answer was: B
When training has completed AND validation has been performed AND results documented AND IV.2 QA/QC has been engaged AND the result is ready to feed IV.6 evaluation
- D4-050 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) go-no-go-gate
Trap: ITERATE to eliminate any difference
Why wrong: Wrong-answer trap — perfect parity may not be achievable.
Correct answer was: B
GO with documented monitoring plan: surface the segment differences in V.4 monitoring, alert on threshold breaches, prepare V.7 contingency for if differences widen post-deployment; engage stakeholders to confirm tolerance is acceptable
- D4-051 · Domain IV · IV.4 — Manage data transformation to conduct data preparation iteration-trigger
Trap: Have the engineer work overtime to finish on time
Why wrong: Wrong-answer trap — overtime without root cause.
Correct answer was: B
Coordinate review of root cause for delay (complexity underestimated, source issues, blocked dependencies, scope creep); engage stakeholders for the impact on downstream phases (IV.5 gate, IV.3 training, schedule); document the decision (extend, descope, or reallocate resources)
- D4-052 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) documentation-required
Trap: The project sponsor alone
Why wrong: Wrong-answer trap — sponsor alone insufficient.
Correct answer was: B
The PM facilitates the documented stakeholder decision; GO is declared via stakeholder sign-off including business owner, technical leadership, security/compliance/privacy where applicable
- D4-053 · Domain IV · IV.1 — Oversee AI/ML model technique(s) documentation-required
Trap: Reject all techniques the PM doesn't understand
Why wrong: Wrong-answer trap — rejection-by-ignorance.
Correct answer was: B
Ensure the data scientist's choice is documented, justified against AI pattern + success criteria, and aligned with operational constraints — but PM doesn't pick the technique
- D4-054 · Domain IV · IV.2 — Oversee AI/ML model QA/QC iteration-trigger
Trap: Note in lessons learned
Why wrong: Wrong-answer trap — note without remediation.
Correct answer was: B
Coordinate root cause: random initialization should produce statistically similar models, not material differences; if material differences exist, investigate seed handling, environment determinism, dependencies; ensure reproducibility is achievable per IV.6 requirement
- D4-055 · Domain IV · IV.5 — Verify data quality (GATE) iteration-trigger
Trap: Have the data scientist remove the proxy feature
Why wrong: Wrong-answer trap — unilateral feature removal without bias evaluation.
Correct answer was: B
Recognize as bias risk (cross-pull I.3); coordinate with bias/fairness review; engage stakeholders to evaluate whether the proxy creates actionable disparate impact; consider feature removal, fairness constraints, or descope; do not proceed without governance review
- D4-056 · Domain IV · IV.3 — Manage AI/ML model training cross-domain-link
Trap: Move to a different cloud provider unilaterally
Why wrong: Wrong-answer trap — unilateral provider change.
Correct answer was: B
Coordinate response: assess maintenance window impact (will training be interrupted, can it resume from checkpoint?); engage data scientist + cloud team; surface options (delay training, use alternative compute, accept interruption with checkpoint resume); document the decision
- D4-057 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) documentation-required
Trap: Just the stakeholders
Why wrong: Wrong-answer trap — partial.
Correct answer was: B
Phase V Go/No-Go Decision Document — performance/bias/robustness/baseline/audit/reproducibility/trustworthy-AI/operational-fit findings; decision (GO/ITERATE/DESCOPE); stakeholder sign-off (multi-team); rationale; downstream actions (Domain V begins)
- D4-058 · Domain IV · IV.4 — Manage data transformation to conduct data preparation iteration-trigger
Trap: Have the engineer choose unilaterally
Why wrong: Wrong-answer trap — engineer-only decision without SME context.
Correct answer was: B
Coordinate with data SMEs (III.2) to evaluate the contradictions: are they data quality issues, real changes over time, or business-meaning differences? Engage stakeholders for the resolution approach (recency, source authority, manual review, exclude); document the decision; do not unilaterally apply recency without SME validation
- D4-059 · Domain IV · IV.2 — Oversee AI/ML model QA/QC documentation-required
Trap: Just the hyperparameters
Why wrong: Wrong-answer trap — partial.
Correct answer was: B
Versions of: code, training data, model artifacts, hyperparameters, evaluation data, environment configuration, dependencies; reproducibility requires versioning all
- D4-060 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) iteration-trigger
Trap: Adjust the model's metrics to match II.8
Why wrong: Wrong-answer trap — unilateral metric adjustment.
Correct answer was: B
ITERATE — performance against criteria isn't meaningfully evaluated. Loop back to II.8 with stakeholders to align metrics, then re-evaluate model against aligned criteria; document both rounds; do not GO with metric-mismatch
- D4-061 · Domain IV · IV.5 — Verify data quality (GATE) documentation-required
Trap: False — IV.5 has no authorization meaning
Why wrong: Wrong-answer trap — IV.5 has explicit authorization.
Correct answer was: B
False — IV.5 GO authorizes IV.3 training to begin. Domain V is authorized by IV.6 GO at end of Phase V (Model Evaluation).
- D4-062 · Domain IV · IV.1 — Oversee AI/ML model technique(s) go-no-go-gate
Trap: Have the ML engineer hot-swap the model
Why wrong: Wrong-answer trap — hot-swap without governance.
Correct answer was: B
Coordinate full process: V.7 contingency for performance issues, V.3 governance for any technique change, IV.1 (not IV.1 again — the original IV.1 is past) requires this is a new model + new IV.6 gate; do not allow mid-production technique change without governance review
- D4-063 · Domain IV · IV.3 — Manage AI/ML model training documentation-required
Trap: Have the PM approve every hyperparameter change
Why wrong: Wrong-answer trap — micro-management.
Correct answer was: B
Hyperparameter tuning is the data scientist's call; PM ensures the search space, methodology, and final values are documented (per IV.2 config management); track tuning runs as part of training progress; ensure final values are reproducible ( criterion)
- D4-064 · Domain IV · IV.5 — Verify data quality (GATE) cross-domain-link
Trap: Have the data scientist proceed with what's available
Why wrong: Wrong-answer trap — proceeding with gaps.
Correct answer was: B
ITERATE — coverage of required attributes is criterion. Dropped attributes mean the data doesn't fully meet III.1 spec. Loop back to III.1 to revise spec OR III.5 to source missing attributes OR descope the model to what's available
- D4-065 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) iteration-trigger
Trap: Have the data scientist add audit trail informally
Why wrong: Wrong-answer trap — informal audit.
Correct answer was: B
ITERATE — Trustworthy-AI alignment is criterion. Coordinate audit-trail completion before GO; engage I.5 stakeholders to confirm audit completeness; do not GO with audit gap
- D4-067 · Domain IV · IV.3 — Manage AI/ML model training iteration-trigger
Trap: Have the data scientist optimize on the fly
Why wrong: Wrong-answer trap — on-the-fly optimization without root cause.
Correct answer was: B
Pause training; coordinate cost analysis (root cause: technique inefficiency, infrastructure mis-sizing, scope creep); engage stakeholders to evaluate options (continue with budget exception, technique change, descope, accept partial training); do not let unilateral cost overrun continue
- D4-068 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) documentation-required
Trap: They're unrelated
Why wrong: Wrong-answer trap — they're tightly coupled.
Correct answer was: B
V.1 (deployment plan) is drafted iteratively before IV.6; IV.6 evaluates operational fit (a criterion); IV.6 GO authorizes V.1 finalization and Domain V execution
- D4-069 · Domain IV · IV.1 — Oversee AI/ML model technique(s) cross-domain-link
Trap: Have the data scientist do quick bias analysis
Why wrong: Wrong-answer trap — quick analysis without governance.
Correct answer was: B
Coordinate bias analysis on fine-tuning data per Domain I.3; foundation model bias inherits to fine-tuned model; ensure analysis is documented and findings inform IV.5/IV.6 gates; engage stakeholders if bias is detected
- D4-070 · Domain IV · IV.4 — Manage data transformation to conduct data preparation documentation-required
Trap: Just the data engineer's notes
Why wrong: Wrong-answer trap — engineer notes aren't a plan.
Correct answer was: B
Transformation rules, pipeline architecture, error handling, ownership of each pipeline stage, reproducibility approach, traceability for QA/QC and audit, integration with III.1 spec and IV.5 gate criteria
- D4-071 · Domain IV · IV.5 — Verify data quality (GATE) documentation-required
Trap: The data steward
Why wrong: Wrong-answer trap — steward provides input.
Correct answer was: B
The PM facilitates the documented stakeholder decision
- D4-072 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) iteration-trigger
Trap: Have the ML engineer adapt the model unilaterally
Why wrong: Wrong-answer trap — unilateral adaptation.
Correct answer was: B
Re-evaluate operational fit ( criterion) against the new infrastructure; if technique no longer fits, ITERATE to IV.1 (technique change) or back to V.1 (infrastructure design); document and engage stakeholders
- D4-073 · Domain IV · IV.3 — Manage AI/ML model training documentation-required
Trap: False — training never completes
Why wrong: Wrong-answer trap — flippant.
Correct answer was: B
False — IV.3 completion = training + validation + documentation + IV.2 QA/QC engagement + ready for IV.6 evaluation
- D4-074 · Domain IV · IV.6 — Verify model ready for operationalization (GATE) documentation-required
Trap: Just inference reproducibility
Why wrong: Wrong-answer trap — same as A.
Correct answer was: B
The training process can be re-run from documented inputs (data version, code version, hyperparameters, environment) and produces a statistically equivalent model; reproducibility extends beyond inference reproducibility to training reproducibility
- D4-075 · Domain IV · IV.5 — Verify data quality (GATE) go-no-go-gate
Trap: GO can be reversed any time
Why wrong: Wrong-answer trap — 'any time' framing diminishes gate weight.
Correct answer was: B
GO authorizes IV.3 training; if subsequent findings (during training, IV.6 evaluation, or production) reveal that data was insufficient, the project may iterate back via subsequent gates; CPMAI is iterative — no gate decision is permanent if new information warrants revisitation
- D5-001 · Domain V · V.1 — Manage creation of AI solution deployment plan oversight-vs-doing
Trap: Have the ML engineer deploy to staging while the plan is written
Why wrong: Wrong-answer trap — partial deployment without plan still violates V.1.
Correct answer was: B
Pause and coordinate creation of the deployment plan before deployment begins, including stakeholder sign-off
- D5-002 · Domain V · V.4 — Oversee AI solution metrics iteration-trigger
Trap: Continue monitoring for another month
Why wrong: Wrong-answer trap — passive monitoring while production degrades.
Correct answer was: B
Investigate root cause (data drift, model decay, scope shift), engage stakeholders, decide between retraining/rolling back/rescoping per the contingency plan
- D5-003 · Domain V · V.2 — Manage AI solution deployment documentation-required
Trap: Deploy → declare success
Why wrong: Wrong-answer trap — runtime alone isn't success.
Correct answer was: C
Verify deployment plan executed → confirm monitoring is live → confirm performance baseline met → confirm governance in place → declare deployment complete
- D5-004 · Domain V · V.3 — Oversee model governance escalation-trigger
Trap: Note it for next quarter's governance review
Why wrong: Wrong-answer trap — passive deferral allows pattern to repeat.
Correct answer was: B
Treat as a governance and accountability incident: validate the deployed version against requirements, document the deviation, escalate per accountability procedures, and reinforce change control
- D5-005 · Domain V · V.7 — Oversee AI solution contingency plan documentation-required
Trap: Defer contingency planning until after deployment to observe failure modes
Why wrong: Wrong-answer trap — observing production for failures isn't planning.
Correct answer was: B
Create the contingency plan now; test it pre-production; document response procedures, owners, escalation paths before going live
- D5-006 · Domain V · V.6 — Manage AI solution transition plan documentation-required
Trap: Email the operations team and let them figure out the handoff
Why wrong: Wrong-answer trap — undocumented handoff loses institutional knowledge.
Correct answer was: B
Coordinate a transition plan including documented artifacts, knowledge transfer, training, escalation paths, and sign-off from the receiving team
- D5-007 · Domain V · V.5 — Prepare final report / lessons learned pattern-recognition
Trap: False — they're only relevant for the same project
Why wrong: Wrong-answer trap — narrow scoping of lessons.
Correct answer was: B
False — CPMAI is iterative. V.5 lessons learned are input to the next iteration's Phase I.
- D5-008 · Domain V · V.3 — Oversee model governance cross-domain-link
Trap: Re-run the prediction and provide the new result
Why wrong: Wrong-answer trap — re-running doesn't reproduce original context.
Correct answer was: B
Provide the audit trail from model governance: input, model version, prediction, timestamp, decision rationale, human-in-the-loop overrides — sourced from the documented governance program
- D5-009 · Domain V · V.4 — Oversee AI solution metrics documentation-required
Trap: Backend / Middleware / Output
Why wrong: Wrong-answer trap — unrelated technical layers.
Correct answer was: B
Business KPIs / Model performance / Operational health
- D5-010 · Domain V · V.1 — Manage creation of AI solution deployment plan documentation-required
Trap: Hot path / Operations / Pipelines / Engineering / Microservices / Streaming
Why wrong: Wrong-answer trap — partial elements.
Correct answer was: A
How served / Operation location / Performance / Escalation / Monitoring / Stakeholder sign-off
- D5-011 · Domain V · V.2 — Manage AI solution deployment go-no-go-gate
Trap: Approve continued deployment
Why wrong: Wrong-answer trap — deployment with monitoring gap is incomplete.
Correct answer was: B
Pause deployment until monitoring is live; deployment success requires monitoring per V.1 plan; do not declare deployment complete with gaps
- D5-012 · Domain V · V.3 — Oversee model governance documentation-required
Trap: Approval / Permission / Authorization / Validation / Backup / Encryption
Why wrong: Wrong-answer trap — security-heavy without versioning/bias.
Correct answer was: B
Access control / Provenance & auditing / Audit logs / Version control / Bias monitoring / Extension controls
- D5-013 · Domain V · V.4 — Oversee AI solution metrics iteration-trigger
Trap: Declare success since technical KPIs are met
Why wrong: Wrong-answer trap — vanity metric trap (technical without business).
Correct answer was: B
Surface the technical-business KPI disconnect; engage stakeholders to investigate (model isn't impacting decisions, predictions aren't acted on, business value definition was wrong); ITERATE may include re-evaluating II.8 success criteria
- D5-014 · Domain V · V.7 — Oversee AI solution contingency plan escalation-trigger
Trap: Apologize to the customer and continue
Why wrong: Wrong-answer trap — apology without containment.
Correct answer was: B
Execute the trustworthy-AI-incident contingency: contain (pause/roll back), audit (document what happened, who affected, cause), notify (stakeholders, regulators per requirements), remediate, document for V.5 lessons learned
- D5-015 · Domain V · V.5 — Prepare final report / lessons learned documentation-required
Trap: Have operations decide when to declare complete
Why wrong: Wrong-answer trap — closeout is PM-coordinated.
Correct answer was: C
Convene Phase VI close-out: document final report and lessons learned, decide on transition to operations or next iteration, capture outstanding risks and dependencies, formal stakeholder sign-off
- D5-018 · Domain V · V.3 — Oversee model governance cross-domain-link
Trap: Have the ML engineer implement A/B testing
Why wrong: Wrong-answer trap — A/B testing implementation isn't governance.
Correct answer was: B
Coordinate documented governance review: A/B routing logic, segment determination, fairness implications (cross-pull I.3), audit trail per segment, accountability for which model serves which user; ensure decision is governed, not technical-default
- D5-019 · Domain V · V.4 — Oversee AI solution metrics documentation-required
Trap: False — drift only affects deep learning models
Why wrong: Wrong-answer trap — drift affects all models.
Correct answer was: B
False — drift is inevitable. Monitoring exists because drift happens. A deployment assuming no drift is the flawed one.
- D5-020 · Domain V · V.6 — Manage AI solution transition plan iteration-trigger
Trap: Have the data scientist remain available indefinitely as backup
Why wrong: Wrong-answer trap — indefinite backup isn't transition completion.
Correct answer was: B
Pause transition; coordinate skill-gap remediation (training, documentation, knowledge transfer sessions, hiring) before transition; engage stakeholders if delay or additional resources are needed
- D5-021 · Domain V · V.5 — Prepare final report / lessons learned documentation-required
Trap: Just the deployment plan
Why wrong: Wrong-answer trap — deployment plan is V.1, not V.5.
Correct answer was: B
Project summary, performance against success criteria (II.8), lessons learned, recommendations for next iteration, outstanding risks and dependencies for ongoing operations
- D5-023 · Domain V · V.1 — Manage creation of AI solution deployment plan cross-domain-link
Trap: Have the ML engineer optimize for CPU
Why wrong: Wrong-answer trap — optimization without IV.1 review.
Correct answer was: B
Recognize cross-domain conflict (IV.6 operational fit + V.1 deployment); engage stakeholders to resolve: provision GPU, reconsider technique (loop IV.1), or descope; do not finalize V.1 with known operational-fit gap
- D5-024 · Domain V · V.4 — Oversee AI solution metrics iteration-trigger
Trap: Have the operations team improve audit-trail capture incrementally
Why wrong: Wrong-answer trap — incremental fix while live operates with audit gap.
Correct answer was: B
Treat as governance/auditability incident (V.3 + I.5 cross-pull); document the gap; engage stakeholders to fix audit-trail capture before further reliance; do not let governance metrics slide because performance metrics are green
- D5-025 · Domain V · V.3 — Oversee model governance documentation-required
Trap: Have the data scientist provide the model
Why wrong: Wrong-answer trap — engineer-only handoff bypasses governance.
Correct answer was: B
Coordinate governance review per V.3 (extension controls in ): assess use case, license, reputation, accountability for resulting model, IP implications; engage legal/compliance/leadership; do not approve external extension without governed process
- D5-026 · Domain V · V.2 — Manage AI solution deployment iteration-trigger
Trap: Continue deployment and patch later
Why wrong: Wrong-answer trap — known vulnerability deployment is governance violation.
Correct answer was: B
Pause deployment; coordinate security/engineering to assess severity; remediate before resuming OR roll forward with documented mitigation if security accepts the risk; do not deploy known-vulnerable dependencies without governance
- D5-027 · Domain V · V.4 — Oversee AI solution metrics documentation-required
Trap: Whenever something breaks
Why wrong: Wrong-answer trap — reactive review misses proactive issues.
Correct answer was: B
Documented in the V.1 deployment plan — typically weekly or monthly with stakeholder access; cadence aligns with metric criticality and project SLAs
- D5-028 · Domain V · V.7 — Oversee AI solution contingency plan documentation-required
Trap: True — documentation alone meets the requirement
Why wrong: Wrong-answer trap — paper compliance.
Correct answer was: B
False — V.7 requires the contingency plan to be tested before production. Untested plans fail in production.
- D5-029 · Domain V · V.1 — Manage creation of AI solution deployment plan go-no-go-gate
Trap: Proceed with deployment and address latency in production
Why wrong: Wrong-answer trap — deferring to production is gate-bypass.
Correct answer was: B
Recognize that IV.6 should have caught operational fit; loop back to IV.6 for re-evaluation OR loop to IV.1 for technique change OR re-architect production environment per V.1; engage stakeholders for the decision
- D5-030 · Domain V · V.3 — Oversee model governance iteration-trigger
Trap: Add bias monitoring at the next major release
Why wrong: Wrong-answer trap — deferred bias remediation = continued exposure.
Correct answer was: B
Treat as governance gap; coordinate immediate addition of bias measurement to monitoring; conduct retrospective bias assessment on production predictions; document the gap and remediation in lessons learned (V.5) and for governance audit (I.5)
- D5-031 · Domain V · V.5 — Prepare final report / lessons learned cross-domain-link
Trap: Have the data scientist explain the shortfall
Why wrong: Wrong-answer trap — explaining doesn't equal documenting.
Correct answer was: B
Document actual vs projected ROI honestly; analyze contributing factors; include lessons learned for next iteration's Phase I (II.5 ROI methodology); engage stakeholders on whether the project's value proposition holds for continuation
- D5-032 · Domain V · V.6 — Manage AI solution transition plan documentation-required
Trap: Just access credentials
Why wrong: Wrong-answer trap — credentials without context.
Correct answer was: B
Documented model artifacts (model file, training pipeline, data lineage), governance plan, monitoring stack, contingency plan, runbooks, escalation paths, audit trail history, and SME contact information
- D5-033 · Domain V · V.4 — Oversee AI solution metrics cross-domain-link
Trap: Approve per-demographic calibration
Why wrong: Wrong-answer trap — well-meaning but governance bypass.
Correct answer was: B
Recognize as bias incident (V.4 + I.3 + I.4): document; engage privacy/legal/ethics; evaluate options (re-source data, retrain with rebalanced data, post-processing fairness, descope minority demographic); do not unilaterally apply per-demographic adjustment without governance and stakeholder review
- D5-034 · Domain V · V.1 — Manage creation of AI solution deployment plan documentation-required
Trap: Just the ML engineer
Why wrong: Wrong-answer trap — engineer execution doesn't equal authorization.
Correct answer was: B
Multiple stakeholders depending on plan content: project sponsor, platform/security, operations, legal/compliance (where applicable), data steward, business owner — coordinated by the PM
- D5-035 · Domain V · V.2 — Manage AI solution deployment oversight-vs-doing
Trap: Have the engineer make the change quickly to keep timeline
Why wrong: Wrong-answer trap — speed-driven approval bypasses governance.
Correct answer was: B
Pause deployment; coordinate architecture-change review per V.1 plan and V.3 governance; engage platform/architecture stakeholders; do not allow architecture changes mid-deployment without governance review; deploy via the documented plan or revise the plan formally
- D5-036 · Domain V · V.7 — Oversee AI solution contingency plan iteration-trigger
Trap: Document the test failure and continue with deployment
Why wrong: Wrong-answer trap — documenting test failure without remediation.
Correct answer was: B
Treat as critical preparation gap; halt deployment; coordinate engineering to fix rollback procedures; re-test before deployment; if deployment is already live, ensure operations team has manual rollback capability and stakeholder awareness of risk
- D5-038 · Domain V · V.4 — Oversee AI solution metrics documentation-required
Trap: What the data scientist thinks is reasonable
Why wrong: Wrong-answer trap — opinion-based baseline isn't governance.
Correct answer was: B
Domain II success criteria (II.8) + IV.2 evaluation results — production metrics tie back to pre-deployment baseline
- D5-039 · Domain V · V.6 — Manage AI solution transition plan phase-identification
Trap: Only when a model is being decommissioned
Why wrong: Wrong-answer trap — decommissioning is one trigger but not exclusive.
Correct answer was: B
Throughout the project — transition plans evolve and execute when team change, vendor change, or successor project requires handoff
- D5-040 · Domain V · V.5 — Prepare final report / lessons learned iteration-trigger
Trap: Suggest the next iteration use a different methodology
Why wrong: Wrong-answer trap — methodology change is overcorrection.
Correct answer was: B
Document explicitly in lessons learned; recommend specific changes to next iteration's Phase I (II.5 ROI methodology — include trustworthy-AI cost; II.9 business case framing); engage stakeholders for organizational learning beyond this project
- D5-041 · Domain V · V.1 — Manage creation of AI solution deployment plan iteration-trigger
Trap: Defer MLOps to a future release
Why wrong: Wrong-answer trap — deferring infrastructure for already-deployed model.
Correct answer was: B
Recognize that MLOps is part of operational fit (IV.6 criterion); engage stakeholders to either add MLOps capability to production environment, change technique to one that fits available MLOps, or descope; do not GO at IV.6 without addressing
- D5-042 · Domain V · V.4 — Oversee AI solution metrics documentation-required
Trap: Technical KPIs are vanity metrics
Why wrong: Wrong-answer trap — partial truth (technical without business is vanity).
Correct answer was: B
Business KPIs measure business outcomes (revenue, cost, CSAT, decision quality); technical KPIs measure model performance (accuracy, F1, recall, latency, throughput); both required, neither alone sufficient
- D5-043 · Domain V · V.7 — Oversee AI solution contingency plan cross-domain-link
Trap: Have the security team handle it
Why wrong: Wrong-answer trap — security alone misses cross-functional response.
Correct answer was: B
Execute the malicious-AI contingency: contain (pause/roll back), audit (24-hour blast radius, affected customers), notify (per regulatory and customer-protection requirements), remediate (patch + adversarial defenses), document for V.5 + governance audit (I.5); coordinate with security/legal/leadership
- D5-044 · Domain V · V.3 — Oversee model governance documentation-required
Trap: False — V.3 is the data scientist's responsibility
Why wrong: Wrong-answer trap — data scientist isn't governance authority.
Correct answer was: B
False — V.3 governance is PM-overseen, organizationally-aligned, multi-team (engineering, security, legal, compliance, data steward, ML engineer); PM coordinates
- D5-045 · Domain V · V.2 — Manage AI solution deployment documentation-required
Trap: Whoever signs the timesheet
Why wrong: Wrong-answer trap — flippant misframing.
Correct answer was: B
The PM, after verifying full success criteria (runtime + monitoring + performance + governance) and documenting completion
- D5-046 · Domain V · V.4 — Oversee AI solution metrics iteration-trigger
Trap: Stop monitoring entirely
Why wrong: Wrong-answer trap — eliminating monitoring is governance violation.
Correct answer was: B
Engage stakeholders to evaluate: monitoring frequency tradeoff (cost vs early detection); examine drift trend (gradual is hard to detect with infrequent monitoring); maintain frequency for high-stakes decisions; document the decision and rationale
- D5-047 · Domain V · V.6 — Manage AI solution transition plan cross-domain-link
Trap: Skip — malicious-AI is operations's concern
Why wrong: Wrong-answer trap — operations needs the plan.
Correct answer was: B
The V.7 contingency plan covering malicious-AI scenarios, monitoring stack for detection, escalation paths, response procedures, periodic test schedule, and stakeholder communication; ensure receiving team is trained on the plan
- D5-048 · Domain V · V.5 — Prepare final report / lessons learned documentation-required
Trap: Include the raw audit trail
Why wrong: Wrong-answer trap — wide distribution of audit data violates governance.
Correct answer was: B
Decline; the audit trail is the input feeding the V.5 report's findings — including raw audit data in the report itself violates audit governance (audit trails are stored securely with controlled access); offer the stakeholder access to the audit system per governance
- D5-049 · Domain V · V.3 — Oversee model governance iteration-trigger
Trap: Document the discrepancy and continue
Why wrong: Wrong-answer trap — documentation without remediation.
Correct answer was: B
Investigate root cause: who deployed unauthorized versions, what's the impact, are versions actively serving users, were governance gaps systemic; coordinate with engineering/operations/security; remediate (decommission + audit + impact assessment); document for V.5 lessons learned
- D5-050 · Domain V · V.7 — Oversee AI solution contingency plan documentation-required
Trap: Whenever an incident occurs
Why wrong: Wrong-answer trap — incident-driven review is reactive.
Correct answer was: B
Documented in the V.7 plan — typically quarterly or aligned with major changes (model update, infrastructure change, regulatory shift); reviewed and tested for effectiveness
- D5-052 · Domain V · V.4 — Oversee AI solution metrics cross-domain-link
Trap: Add fairness measurement at next major release
Why wrong: Wrong-answer trap — deferring fairness measurement.
Correct answer was: B
Treat as cross-domain governance gap (II.8 + I.3 + V.4); coordinate immediate addition of fairness metric; conduct retrospective fairness assessment on production predictions; document the gap (V.5 lessons learned + I.5 audit) and remediate; do not let success-criteria metrics slip from monitoring
- D5-053 · Domain V · V.3 — Oversee model governance cross-domain-link
Trap: Have the data scientist add post-hoc explanations
Why wrong: Wrong-answer trap — XAI may not satisfy regulator's bar.
Correct answer was: B
Coordinate with legal/compliance to assess regulation requirements; engage data science to add XAI capability OR consider model replacement (loop IV.1) if XAI is insufficient; update governance plan; engage stakeholders on schedule/scope/cost impact
- D5-054 · Domain V · V.2 — Manage AI solution deployment iteration-trigger
Trap: Have the engineer add error handling
Why wrong: Wrong-answer trap — fix without root-cause may mask issue.
Correct answer was: B
Coordinate root-cause investigation: is the error in input validation, model code, or upstream pipeline? Is the 5% within or beyond contingency tolerance? Engage stakeholders for remediation decision; do not declare deployment fully successful with documented error rate
- D5-055 · Domain V · V.6 — Manage AI solution transition plan documentation-required
Trap: True — they should never need help
Why wrong: Wrong-answer trap — black-or-white framing.
Correct answer was: B
Mostly true — the receiving team should be capable of independent operation, but documented escalation paths back to the project team or successor SMEs are still part of the transition plan
- D5-056 · Domain V · V.5 — Prepare final report / lessons learned documentation-required
Trap: Hire a project manager
Why wrong: Wrong-answer trap — flippant.
Correct answer was: B
Specific actionable recommendations: stakeholder communication plan (cadence, format, audience-specific framing), documented decision criteria for what reaches leadership vs operates internally, training for the team on stakeholder management, integration into next iteration's project plan
- D5-057 · Domain V · V.4 — Oversee AI solution metrics iteration-trigger
Trap: Continue and let production discover the issue
Why wrong: Wrong-answer trap — passive monitoring.
Correct answer was: B
Recognize as Domain II + Domain V cross-pull; engage stakeholders to revisit success criteria (II.8) and possibly business problem (Phase I); the model may be working correctly per its training but the business has moved; iterate Phase I/II if needed
- D5-058 · Domain V · V.7 — Oversee AI solution contingency plan pattern-recognition
Trap: Audit (document what happened)
Why wrong: Wrong-answer trap — audit is third.
Correct answer was: B
Detect (monitoring surfaces the issue)
- D5-059 · Domain V · V.3 — Oversee model governance documentation-required
Trap: Just the model version
Why wrong: Wrong-answer trap — version alone.
Correct answer was: B
Input data, model version, prediction, timestamp, decision rationale, human-in-the-loop overrides
- D5-060 · Domain V · V.1 — Manage creation of AI solution deployment plan iteration-trigger
Trap: Have the engineer find a workaround
Why wrong: Wrong-answer trap — engineer workaround without documented decision.
Correct answer was: B
Engage stakeholders to evaluate options: scope to batch-acceptable use case (technique change loop to IV.1), invest in specialized services (cost decision), find alternative cloud, or descope; document the constraint for next iteration's Phase I
- D5-061 · Domain V · V.2 — Manage AI solution deployment documentation-required
Trap: Have the data scientist write the change documentation while deployment continues
Why wrong: Wrong-answer trap — concurrent change without governance review.
Correct answer was: B
Pause to coordinate plan revision: document the proposed change, evaluate impact, get stakeholder sign-off on the revision, then either resume per revised plan or document the proposal for next iteration; do not deviate from documented plan without governance
- D5-062 · Domain V · V.6 — Manage AI solution transition plan iteration-trigger
Trap: Negotiate to transition as-is and let receiving team fix later
Why wrong: Wrong-answer trap — passing debt without agreement.
Correct answer was: B
Recognize as serious finding requiring V.5 lessons learned + V.3 governance remediation; coordinate gap closure before transition (or with documented receiving-team agreement on remediation timeline); do not transition known governance debt without explicit arrangement
- D5-063 · Domain V · V.5 — Prepare final report / lessons learned documentation-required
Trap: True — wins are obvious
Why wrong: Wrong-answer trap — implicit assumption.
Correct answer was: B
False — V.5 captures both successes and failures; what worked is as important as what didn't, since both inform next iteration's Phase I
- D5-064 · Domain V · V.4 — Oversee AI solution metrics documentation-required
Trap: Nothing — usage data is privacy-sensitive
Why wrong: Wrong-answer trap — privacy concerns are valid but don't eliminate monitoring.
Correct answer was: B
Who is using the model, how, how often — useful for training future versions, governance, and detecting anomalous usage
- D5-065 · Domain V · V.7 — Oversee AI solution contingency plan iteration-trigger
Trap: Continue while retraining happens in background
Why wrong: Wrong-answer trap — degraded model serving production.
Correct answer was: B
Recognize as data-drift incident requiring contingency assessment: evaluate impact on predictions, decide whether to roll back to last-known-good while retraining, retrain with appropriate validation, govern the new version per V.3, monitor closely (V.4); engage stakeholders for the rollback-during-retrain decision
- D5-066 · Domain V · V.3 — Oversee model governance phase-identification
Trap: I.5 is for projects; V.3 is for products
Why wrong: Wrong-answer trap — false dichotomy.
Correct answer was: B
Domain I.5 defines the principle; Domain V.3 is the production execution. V.3's audit logs and accountability documentation are how I.5's principle is operationalized.
- D5-067 · Domain V · V.1 — Manage creation of AI solution deployment plan documentation-required
Trap: After IV.6 GO authorizes Domain V work
Why wrong: Wrong-answer trap — finalize after IV.6 but draft before.
Correct answer was: D
Earlier — drafting can begin in Phase I and refine through Phase IV/V; many deployment-plan elements (operational fit, monitoring, contingency) need to inform earlier decisions; finalization comes after IV.6 GO but draft is ongoing
- D5-068 · Domain V · V.2 — Manage AI solution deployment cross-domain-link
Trap: Defer hardening to next release
Why wrong: Wrong-answer trap — deferring security.
Correct answer was: B
Pause deployment; coordinate security/platform/engineering to harden environment per organizational standards before resuming; this is V.1 plan compliance + Domain I.1 cross-pull (security plan)
- D5-070 · Domain V · V.5 — Prepare final report / lessons learned iteration-trigger
Trap: Skip — the risk is resolved
Why wrong: Wrong-answer trap — resolution doesn't end the learning.
Correct answer was: B
Document: original risk assessment, what materialized, how V.7 contingency addressed it, what worked well, recommendations for similar risks in next iteration's II.3 + V.7
- D5-071 · Domain V · V.4 — Oversee AI solution metrics documentation-required
Trap: Just the data scientist has access
Why wrong: Wrong-answer trap — narrow access defeats stakeholder visibility.
Correct answer was: B
Stakeholder access defined per the V.1 plan; documented review cadence; alerting on threshold breaches; tied back to Domain II success criteria; integrated with audit (V.3 / I.5) where regulatory
- D5-072 · Domain V · V.7 — Oversee AI solution contingency plan documentation-required
Trap: Just security
Why wrong: Wrong-answer trap — security alone misses other roles.
Correct answer was: B
Cross-functional: engineering (rollback execution), operations (incident response), security (incident-response), stakeholders (decision-making during incident), receiving team (if transition is upcoming) — coordinated by PM
- D5-073 · Domain V · V.3 — Oversee model governance iteration-trigger
Trap: Trust performance and skip re-evaluation
Why wrong: Wrong-answer trap — trusting current state ignores drift.
Correct answer was: B
Schedule documented re-evaluation per V.3 governance; even well-performing models need periodic validation; bias measurements may have shifted; data drift may be slow; reproducibility may be at risk; document the re-evaluation outcome
- D5-074 · Domain V · V.1 — Manage creation of AI solution deployment plan cross-domain-link
Trap: Proceed with cross-jurisdiction inference
Why wrong: Wrong-answer trap — knowing privacy plan violation.
Correct answer was: B
Recognize as cross-domain conflict (V.1 + I.1 + I.4); coordinate with privacy/compliance/legal; engage stakeholders to resolve: deploy model in compliant jurisdiction, redesign data flow to keep data in jurisdiction, update privacy plan if alternative is acceptable, or descope the geography; do not finalize V.1 with known privacy plan violation
- D5-075 · Domain V · V.2 — Manage AI solution deployment documentation-required
Trap: True — error-free serving is the goal
Why wrong: Wrong-answer trap — narrow technical framing.
Correct answer was: B
False — deployment success requires runtime + monitoring + performance against criteria + governance in place; runtime alone is necessary but insufficient