IAPP AIGP Practice Test 68
Question 670
A speech model trained on one accent performs poorly for customers speaking regional dialects. Model assurance reviewers need a documented control before approving continued experimentation. Which specific implementation best addresses the identified gap?
- Record end-to-end feature lineage and review proxy risks in engineered attributes.
- Validate explanation methods and document what attributions do and do not establish.
- Improve approved data coverage and labeling with a targeted, lawful sampling plan.
- Define and refine labeling policy with expert adjudication and representative agreement testing.
Explanation
Correct Answer
C:Improve approved data coverage and labeling with a targeted, lawful sampling plan. Correct. It directly reduces or governs the specific documented risk. Training data quality includes representativeness for the intended use population.
Incorrect Answers
A:Record end-to-end feature lineage and review proxy risks in engineered attributes. Not selected. That action treats another control objective, not the specific gap described. It belongs with Data lineage for transformations: Lineage tracks how raw data becomes model-ready features, including indirect sensitive signals. The present case instead concerns Data quality and representativeness.
B:Validate explanation methods and document what attributions do and do not establish. Not selected. That action treats another control objective, not the specific gap described. It belongs with Interpretability testing: An explanation tool may be misleading unless its reliability and limitations are evaluated. The present case instead concerns Data quality and representativeness.
D:Define and refine labeling policy with expert adjudication and representative agreement testing. Not selected. That action treats another control objective, not the specific gap described. It belongs with Label quality and inter-rater agreement: Ground-truth quality shapes model behavior and can hide systematic harm. The present case instead concerns Data quality and representativeness.
Question 671
A speech model trained on one accent performs poorly for customers speaking regional dialects. Model assurance reviewers need a documented control before approving continued experimentation. Which acceptance test most directly verifies the relevant control?
- Recreate selected features from source data and confirm sensitive transformations are visible.
- Evaluate both error differences and selection effects on representative groups after mitigation.
- Run a held-out dialect evaluation and compare false-negative rates and uncertainty thresholds.
- Test for leakage indicators and compare results on a genuinely unseen time period.
Explanation
Correct Answer
C:Run a held-out dialect evaluation and compare false-negative rates and uncertainty thresholds. Correct. It yields measurable evidence that the case-specific safeguard works. Training data quality includes representativeness for the intended use population.
Incorrect Answers
A:Recreate selected features from source data and confirm sensitive transformations are visible. Not selected. That test would not demonstrate this particular risk has been addressed. It belongs with Data lineage for transformations: Lineage tracks how raw data becomes model-ready features, including indirect sensitive signals. The present case instead concerns Data quality and representativeness.
B:Evaluate both error differences and selection effects on representative groups after mitigation. Not selected. That test would not demonstrate this particular risk has been addressed. It belongs with Bias measurement and mitigation: Historic training labels can reproduce inequity even when protected fields are omitted. The present case instead concerns Data quality and representativeness.
D:Test for leakage indicators and compare results on a genuinely unseen time period. Not selected. That test would not demonstrate this particular risk has been addressed. It belongs with Dataset leakage and temporal splits: A contaminated test set can give convincing but invalid evidence of generalization. The present case instead concerns Data quality and representativeness.
Question 672
A speech model trained on one accent performs poorly for customers speaking regional dialects. Model assurance reviewers need a documented control before approving continued experimentation. For a defensible decision, choose TWO: one diagnosis and one controlled response.
- Define and refine labeling policy with expert adjudication and representative agreement testing.
- Measure sample coverage, label consistency, subgroup representation and production-population mismatch.
- Assess explanation stability, fidelity and whether the method matches model and use-case limitations.
- Improve approved data coverage and labeling with a targeted, lawful sampling plan.
Explanation
Correct Answers
B:Measure sample coverage, label consistency, subgroup representation and production-population mismatch. Correct. As the first, evidence-gathering half of the requested pair, this establishes the specific risk boundary. Training data quality includes representativeness for the intended use population.
D:Improve approved data coverage and labeling with a targeted, lawful sampling plan. Correct. As the second, implementation half of the requested pair, this directly applies the needed safeguard. Training data quality includes representativeness for the intended use population.
Incorrect Answers
A:Define and refine labeling policy with expert adjudication and representative agreement testing. Not selected. It addresses a different governance mechanism, not the required assessment-and-control pair. It belongs with Label quality and inter-rater agreement: Ground-truth quality shapes model behavior and can hide systematic harm. The present case instead concerns Data quality and representativeness.
C:Assess explanation stability, fidelity and whether the method matches model and use-case limitations. Not selected. It addresses a different governance mechanism, not the required assessment-and-control pair. It belongs with Interpretability testing: An explanation tool may be misleading unless its reliability and limitations are evaluated. The present case instead concerns Data quality and representativeness.
Question 673
A speech model trained on one accent performs poorly for customers speaking regional dialects. A material training or architecture change makes the previous technical approval insufficient. Which diagnostic task establishes the risk relevant to this described system?
- Select foreseeable corruption, adversarial perturbation and distribution-shift scenarios to test.
- Measure sample coverage, label consistency, subgroup representation and production-population mismatch.
- Trace raw data, feature engineering, joins and transformations to the final training inputs.
- Audit labeling guidance, disagreement patterns, annotator context and ambiguous categories.
Explanation
Correct Answer
B:Measure sample coverage, label consistency, subgroup representation and production-population mismatch. Correct. It establishes the relevant factual risk boundary before implementation. Training data quality includes representativeness for the intended use population.
Incorrect Answers
A:Select foreseeable corruption, adversarial perturbation and distribution-shift scenarios to test. Not selected. That check primarily diagnoses a different governance issue. It belongs with Adversarial robustness testing: Security and robustness tests must cover plausible operating and adversarial conditions. The present case instead concerns Data quality and representativeness.
C:Trace raw data, feature engineering, joins and transformations to the final training inputs. Not selected. That check primarily diagnoses a different governance issue. It belongs with Data lineage for transformations: Lineage tracks how raw data becomes model-ready features, including indirect sensitive signals. The present case instead concerns Data quality and representativeness.
D:Audit labeling guidance, disagreement patterns, annotator context and ambiguous categories. Not selected. That check primarily diagnoses a different governance issue. It belongs with Label quality and inter-rater agreement: Ground-truth quality shapes model behavior and can hide systematic harm. The present case instead concerns Data quality and representativeness.
Question 674
A speech model trained on one accent performs poorly for customers speaking regional dialects. A material training or architecture change makes the previous technical approval insufficient. What is the strongest proportionate corrective design decision?
- Add robustness tests, input controls and fallback behavior for unsafe predictions.
- Improve approved data coverage and labeling with a targeted, lawful sampling plan.
- Define and refine labeling policy with expert adjudication and representative agreement testing.
- Apply legally appropriate sampling, fairness evaluation and mitigation with job-related validation.
Explanation
Correct Answer
B:Improve approved data coverage and labeling with a targeted, lawful sampling plan. Correct. It directly reduces or governs the specific documented risk. Training data quality includes representativeness for the intended use population.
Incorrect Answers
A:Add robustness tests, input controls and fallback behavior for unsafe predictions. Not selected. That action treats another control objective, not the specific gap described. It belongs with Adversarial robustness testing: Security and robustness tests must cover plausible operating and adversarial conditions. The present case instead concerns Data quality and representativeness.
C:Define and refine labeling policy with expert adjudication and representative agreement testing. Not selected. That action treats another control objective, not the specific gap described. It belongs with Label quality and inter-rater agreement: Ground-truth quality shapes model behavior and can hide systematic harm. The present case instead concerns Data quality and representativeness.
D:Apply legally appropriate sampling, fairness evaluation and mitigation with job-related validation. Not selected. That action treats another control objective, not the specific gap described. It belongs with Bias measurement and mitigation: Historic training labels can reproduce inequity even when protected fields are omitted. The present case instead concerns Data quality and representativeness.
Question 675
A speech model trained on one accent performs poorly for customers speaking regional dialects. A material training or architecture change makes the previous technical approval insufficient. Which observable evidence best demonstrates the chosen safeguard is working?
- Measure agreement and error on a reviewed gold-standard sample before training.
- Re-run an approved benchmark and compare results within documented tolerances.
- Evaluate both error differences and selection effects on representative groups after mitigation.
- Run a held-out dialect evaluation and compare false-negative rates and uncertainty thresholds.
Explanation
Correct Answer
D:Run a held-out dialect evaluation and compare false-negative rates and uncertainty thresholds. Correct. It yields measurable evidence that the case-specific safeguard works. Training data quality includes representativeness for the intended use population.
Incorrect Answers
A:Measure agreement and error on a reviewed gold-standard sample before training. Not selected. That test would not demonstrate this particular risk has been addressed. It belongs with Label quality and inter-rater agreement: Ground-truth quality shapes model behavior and can hide systematic harm. The present case instead concerns Data quality and representativeness.
B:Re-run an approved benchmark and compare results within documented tolerances. Not selected. That test would not demonstrate this particular risk has been addressed. It belongs with Testing evidence and reproducibility: Reproducibility makes AI testing evidence reviewable rather than anecdotal. The present case instead concerns Data quality and representativeness.
C:Evaluate both error differences and selection effects on representative groups after mitigation. Not selected. That test would not demonstrate this particular risk has been addressed. It belongs with Bias measurement and mitigation: Historic training labels can reproduce inequity even when protected fields are omitted. The present case instead concerns Data quality and representativeness.
Question 676
A speech model trained on one accent performs poorly for customers speaking regional dialects. A material training or architecture change makes the previous technical approval insufficient. A material exception remains unresolved. Which containment decision best protects affected parties?
- Provide accessible non-AI service paths while underrepresented speech performance is corrected.
- Remove problematic derived features and invalidate affected model releases when lineage cannot be verified.
- Quarantine undocumented data and retrain or re-evaluate models if unauthorized material was used.
- Inspect collection origin, licenses, consent where applicable, lineage and transformation history.
Explanation
Correct Answer
A:Provide accessible non-AI service paths while underrepresented speech performance is corrected. Correct. It limits further reliance or exposure while the risk remains unresolved. Training data quality includes representativeness for the intended use population.
Incorrect Answers
B:Remove problematic derived features and invalidate affected model releases when lineage cannot be verified. Not selected. That response contains a different failure mode and leaves the stated risk unresolved. It belongs with Data lineage for transformations: Lineage tracks how raw data becomes model-ready features, including indirect sensitive signals. The present case instead concerns Data quality and representativeness.
C:Quarantine undocumented data and retrain or re-evaluate models if unauthorized material was used. Not selected. That response contains a different failure mode and leaves the stated risk unresolved. It belongs with Training data provenance: Data provenance establishes what entered training and under what conditions. The present case instead concerns Data quality and representativeness.
D:Inspect collection origin, licenses, consent where applicable, lineage and transformation history. Not selected. This action relates to Training data provenance. Data provenance establishes what entered training and under what conditions. For the contain decision, it does not resolve the specifically described issue of Data quality and representativeness in A speech model trained on one accent performs poorly for customers speaking regional dialects. A material training or architecture change makes the previous technical approval insufficient.
Question 677
A forecasting team randomly splits transaction records even though near-identical customer sequences appear in training and test partitions. The development team must show evidence before promoting the next model version. Which diagnostic task establishes the risk relevant to this described system?
- Identify missing code, data, parameter, environment and test-set versions.
- Identify how sample selection, labels and historical practices may propagate protected-group disadvantages.
- Check entity, time and duplicate overlap that may leak future information or memorized records.
- Measure sample coverage, label consistency, subgroup representation and production-population mismatch.
Explanation
Correct Answer
C:Check entity, time and duplicate overlap that may leak future information or memorized records. Correct. It establishes the relevant factual risk boundary before implementation. A contaminated test set can give convincing but invalid evidence of generalization.
Incorrect Answers
A:Identify missing code, data, parameter, environment and test-set versions. Not selected. That check primarily diagnoses a different governance issue. It belongs with Testing evidence and reproducibility: Reproducibility makes AI testing evidence reviewable rather than anecdotal. The present case instead concerns Dataset leakage and temporal splits.
B:Identify how sample selection, labels and historical practices may propagate protected-group disadvantages. Not selected. That check primarily diagnoses a different governance issue. It belongs with Bias measurement and mitigation: Historic training labels can reproduce inequity even when protected fields are omitted. The present case instead concerns Dataset leakage and temporal splits.
D:Measure sample coverage, label consistency, subgroup representation and production-population mismatch. Not selected. That check primarily diagnoses a different governance issue. It belongs with Data quality and representativeness: Training data quality includes representativeness for the intended use population. The present case instead concerns Dataset leakage and temporal splits.
Question 678
A forecasting team randomly splits transaction records even though near-identical customer sequences appear in training and test partitions. The development team must show evidence before promoting the next model version. What is the strongest proportionate corrective design decision?
- Validate explanation methods and document what attributions do and do not establish.
- Use temporal or group-aware splits and isolate evaluation records from training and tuning.
- Record end-to-end feature lineage and review proxy risks in engineered attributes.
- Maintain reproducible experiments and protected evaluation artifacts with traceable approvals.
Explanation
Correct Answer
B:Use temporal or group-aware splits and isolate evaluation records from training and tuning. Correct. It directly reduces or governs the specific documented risk. A contaminated test set can give convincing but invalid evidence of generalization.
Incorrect Answers
A:Validate explanation methods and document what attributions do and do not establish. Not selected. That action treats another control objective, not the specific gap described. It belongs with Interpretability testing: An explanation tool may be misleading unless its reliability and limitations are evaluated. The present case instead concerns Dataset leakage and temporal splits.
C:Record end-to-end feature lineage and review proxy risks in engineered attributes. Not selected. That action treats another control objective, not the specific gap described. It belongs with Data lineage for transformations: Lineage tracks how raw data becomes model-ready features, including indirect sensitive signals. The present case instead concerns Dataset leakage and temporal splits.
D:Maintain reproducible experiments and protected evaluation artifacts with traceable approvals. Not selected. That action treats another control objective, not the specific gap described. It belongs with Testing evidence and reproducibility: Reproducibility makes AI testing evidence reviewable rather than anecdotal. The present case instead concerns Dataset leakage and temporal splits.
Question 679
A forecasting team randomly splits transaction records even though near-identical customer sequences appear in training and test partitions. The development team must show evidence before promoting the next model version. Which observable evidence best demonstrates the chosen safeguard is working?
- Test for leakage indicators and compare results on a genuinely unseen time period.
- Measure agreement and error on a reviewed gold-standard sample before training.
- Re-run an approved benchmark and compare results within documented tolerances.
- Test similar inputs for explanation consistency and verify reviewers do not treat correlates as proven causes.
Explanation
Correct Answer
A:Test for leakage indicators and compare results on a genuinely unseen time period. Correct. It yields measurable evidence that the case-specific safeguard works. A contaminated test set can give convincing but invalid evidence of generalization.
Incorrect Answers
B:Measure agreement and error on a reviewed gold-standard sample before training. Not selected. That test would not demonstrate this particular risk has been addressed. It belongs with Label quality and inter-rater agreement: Ground-truth quality shapes model behavior and can hide systematic harm. The present case instead concerns Dataset leakage and temporal splits.
C:Re-run an approved benchmark and compare results within documented tolerances. Not selected. That test would not demonstrate this particular risk has been addressed. It belongs with Testing evidence and reproducibility: Reproducibility makes AI testing evidence reviewable rather than anecdotal. The present case instead concerns Dataset leakage and temporal splits.
D:Test similar inputs for explanation consistency and verify reviewers do not treat correlates as proven causes. Not selected. That test would not demonstrate this particular risk has been addressed. It belongs with Interpretability testing: An explanation tool may be misleading unless its reliability and limitations are evaluated. The present case instead concerns Dataset leakage and temporal splits.
Question 680
A forecasting team randomly splits transaction records even though near-identical customer sequences appear in training and test partitions. The development team must show evidence before promoting the next model version. A material exception remains unresolved. Which containment decision best protects affected parties?
- Remove problematic derived features and invalidate affected model releases when lineage cannot be verified.
- Discard contaminated benchmark claims and rerun controlled evaluation before release.
- Withdraw unverifiable performance claims and rerun controlled tests before launch.
- Switch to a validated safe mode when input quality falls outside the assessed operating domain.
Explanation
Correct Answer
B:Discard contaminated benchmark claims and rerun controlled evaluation before release. Correct. It limits further reliance or exposure while the risk remains unresolved. A contaminated test set can give convincing but invalid evidence of generalization.
Incorrect Answers
A:Remove problematic derived features and invalidate affected model releases when lineage cannot be verified. Not selected. That response contains a different failure mode and leaves the stated risk unresolved. It belongs with Data lineage for transformations: Lineage tracks how raw data becomes model-ready features, including indirect sensitive signals. The present case instead concerns Dataset leakage and temporal splits.
C:Withdraw unverifiable performance claims and rerun controlled tests before launch. Not selected. That response contains a different failure mode and leaves the stated risk unresolved. It belongs with Testing evidence and reproducibility: Reproducibility makes AI testing evidence reviewable rather than anecdotal. The present case instead concerns Dataset leakage and temporal splits.
D:Switch to a validated safe mode when input quality falls outside the assessed operating domain. Not selected. That response contains a different failure mode and leaves the stated risk unresolved. It belongs with Adversarial robustness testing: Security and robustness tests must cover plausible operating and adversarial conditions. The present case instead concerns Dataset leakage and temporal splits.