Exam Bundle

The monitoring report labels distribution shift and actual error metrics distinctly and does not infer accuracy loss from drift alone. This would verify a different operation: Distinguish covariate drift from verified performance degradation and evaluate outcome labels and model-quality metrics separately. That separate action addresses: An operations dashboard treats a rising model input drift score as proof that prediction correctness already decreased without label evidence. It cannot establish the required Bluehaven Transit correction for: S3 inference request captures accumulate indefinitely in a costly storage class even after their approved monitoring retention period expires. The decisive acceptance check is instead: Objects past the retention window transition or expire as intended while protected evidence remains retrievable.

The production role loads the shared artifact and a separate untrusted account still receives an access denial. This would verify a different operation: Configure precise cross-account package, S3 and KMS key permissions for the trusted production consumer role. That separate action addresses: A cross-account model registry package is shared with a production account but the referenced encrypted S3 model artifact cannot be decrypted there. It cannot establish the required Bluehaven Transit correction for: S3 inference request captures accumulate indefinitely in a costly storage class even after their approved monitoring retention period expires. The decisive acceptance check is instead: Objects past the retention window transition or expire as intended while protected evidence remains retrievable.

A simulated stale feature feed triggers a dependency alarm even though the SageMaker endpoint remains available. This would verify a different operation: Monitor end-to-end feature freshness, dependency errors and successful business prediction rates in addition to endpoint health. That separate action addresses: Production predictions fail silently during a partial downstream feature outage while endpoint uptime remains normal. It cannot establish the required Bluehaven Transit correction for: S3 inference request captures accumulate indefinitely in a costly storage class even after their approved monitoring retention period expires. The decisive acceptance check is instead: Objects past the retention window transition or expire as intended while protected evidence remains retrievable.

Version monitoring baselines with each released model and capture a reviewed transition policy between old and new expectations. It does not directly address the Northgate Telecom case: A model endpoint scales to zero immediately after every request but subsequent cold starts repeatedly violate the low-latency objective. The relevant acceptance criterion for this case is: A low-traffic replay meets p95 response requirements and documents the cost of maintaining necessary warm capacity.

Monitor inference request schema and null-value rates against a versioned accepted contract before treating the endpoint as healthy. It does not directly address the Northgate Telecom case: A model endpoint scales to zero immediately after every request but subsequent cold starts repeatedly violate the low-latency objective. The relevant acceptance criterion for this case is: A low-traffic replay meets p95 response requirements and documents the cost of maintaining necessary warm capacity.

Use appropriate SageMaker Clarify post-training bias measures and valid governed outcome data with documented fairness interpretation. It does not directly address the Northgate Telecom case: A model endpoint scales to zero immediately after every request but subsequent cold starts repeatedly violate the low-latency objective. The relevant acceptance criterion for this case is: A low-traffic replay meets p95 response requirements and documents the cost of maintaining necessary warm capacity.

The monitoring report labels distribution shift and actual error metrics distinctly and does not infer accuracy loss from drift alone. This would verify a different operation: Distinguish covariate drift from verified performance degradation and evaluate outcome labels and model-quality metrics separately. That separate action addresses: An operations dashboard treats a rising model input drift score as proof that prediction correctness already decreased without label evidence. It cannot establish the required Northgate Telecom correction for: A model endpoint scales to zero immediately after every request but subsequent cold starts repeatedly violate the low-latency objective. The decisive acceptance check is instead: A low-traffic replay meets p95 response requirements and documents the cost of maintaining necessary warm capacity.

A controlled prediction from the new variant appears in the correct monitoring dataset and the old variant is not mistaken for current. This would verify a different operation: Update monitored endpoint or variant configuration and baseline references as part of the versioned deployment process. That separate action addresses: A risk model is deployed with a new model version but its SageMaker Model Monitor schedule continues capturing only the old endpoint variant. It cannot establish the required Northgate Telecom correction for: A model endpoint scales to zero immediately after every request but subsequent cold starts repeatedly violate the low-latency objective. The decisive acceptance check is instead: A low-traffic replay meets p95 response requirements and documents the cost of maintaining necessary warm capacity.

The inference container accesses authorized model artifacts privately and an external egress test remains blocked. This would verify a different operation: Use approved private subnets, security groups and VPC endpoints where supported to restrict unnecessary network egress. That separate action addresses: A production inference container is placed in subnets with internet egress although its required datasets and dependencies are private AWS services. It cannot establish the required Northgate Telecom correction for: A model endpoint scales to zero immediately after every request but subsequent cold starts repeatedly violate the low-latency objective. The decisive acceptance check is instead: A low-traffic replay meets p95 response requirements and documents the cost of maintaining necessary warm capacity.

Scope CloudWatch log actions to required resource ARNs and separate log configuration permissions from ordinary invocation needs. It does not directly address the Harborline Logistics case: A model endpoint scales to zero immediately after every request but subsequent cold starts repeatedly violate the low-latency objective. The relevant acceptance criterion for this case is: A low-traffic replay meets p95 response requirements and documents the cost of maintaining necessary warm capacity.

Version monitoring baselines with each released model and capture a reviewed transition policy between old and new expectations. It does not directly address the Harborline Logistics case: A model endpoint scales to zero immediately after every request but subsequent cold starts repeatedly violate the low-latency objective. The relevant acceptance criterion for this case is: A low-traffic replay meets p95 response requirements and documents the cost of maintaining necessary warm capacity.

Run shadow or controlled A/B inference comparisons with logged model version and matched request IDs before full rollout. It does not directly address the Harborline Logistics case: A model endpoint scales to zero immediately after every request but subsequent cold starts repeatedly violate the low-latency objective. The relevant acceptance criterion for this case is: A low-traffic replay meets p95 response requirements and documents the cost of maintaining necessary warm capacity.

The monitoring report labels distribution shift and actual error metrics distinctly and does not infer accuracy loss from drift alone. This would verify a different operation: Distinguish covariate drift from verified performance degradation and evaluate outcome labels and model-quality metrics separately. That separate action addresses: An operations dashboard treats a rising model input drift score as proof that prediction correctness already decreased without label evidence. It cannot establish the required Harborline Logistics correction for: A model endpoint scales to zero immediately after every request but subsequent cold starts repeatedly violate the low-latency objective. The decisive acceptance check is instead: A low-traffic replay meets p95 response requirements and documents the cost of maintaining necessary warm capacity.

A controlled malformed-input batch is counted in the rejection metric and overall error reporting includes all attempted records. This would verify a different operation: Track total received records, rejected records and successful predictions with clear denominators across batch processing. That separate action addresses: A batch prediction system drops malformed records but the monitoring report tracks only successfully scored inputs and understates total error rate. It cannot establish the required Harborline Logistics correction for: A model endpoint scales to zero immediately after every request but subsequent cold starts repeatedly violate the low-latency objective. The decisive acceptance check is instead: A low-traffic replay meets p95 response requirements and documents the cost of maintaining necessary warm capacity.

A controlled event's original prediction joins to the matching eventual label without exposing unauthorized personal identifiers. This would verify a different operation: Retain privacy-preserving stable correlation keys through prediction capture and delayed outcome ingestion. That separate action addresses: An inference monitoring workflow cannot associate captured predictions with later labels because request identifiers are stripped during processing. It cannot establish the required Harborline Logistics correction for: A model endpoint scales to zero immediately after every request but subsequent cold starts repeatedly violate the low-latency objective. The decisive acceptance check is instead: A low-traffic replay meets p95 response requirements and documents the cost of maintaining necessary warm capacity.

Create a SageMaker Model Monitor data-quality baseline and schedule inference feature-distribution checks against it. It does not directly address the Silvercrest University case: An enterprise ML team assumes a larger GPU instance always reduces cost although the model is small and memory bandwidth is not a bottleneck. The relevant acceptance criterion for this case is: The selected instance meets the latency requirement with measured lower normalized cost than the oversized GPU option.

Retain privacy-preserving stable correlation keys through prediction capture and delayed outcome ingestion. It does not directly address the Silvercrest University case: An enterprise ML team assumes a larger GPU instance always reduces cost although the model is small and memory bandwidth is not a bottleneck. The relevant acceptance criterion for this case is: The selected instance meets the latency requirement with measured lower normalized cost than the oversized GPU option.

Run shadow or controlled A/B inference comparisons with logged model version and matched request IDs before full rollout. It does not directly address the Silvercrest University case: An enterprise ML team assumes a larger GPU instance always reduces cost although the model is small and memory bandwidth is not a bottleneck. The relevant acceptance criterion for this case is: The selected instance meets the latency requirement with measured lower normalized cost than the oversized GPU option.

A controlled malformed-input batch is counted in the rejection metric and overall error reporting includes all attempted records. This would verify a different operation: Track total received records, rejected records and successful predictions with clear denominators across batch processing. That separate action addresses: A batch prediction system drops malformed records but the monitoring report tracks only successfully scored inputs and understates total error rate. It cannot establish the required Silvercrest University correction for: An enterprise ML team assumes a larger GPU instance always reduces cost although the model is small and memory bandwidth is not a bottleneck. The decisive acceptance check is instead: The selected instance meets the latency requirement with measured lower normalized cost than the oversized GPU option.

An authorized monitoring sample supports drift analysis and a security scan finds no forbidden raw identifiers in captured payloads. This would verify a different operation: Configure bounded and redacted inference data capture with access controls, sampling and retention matched to monitoring requirements. That separate action addresses: A SageMaker endpoint stores all prediction payloads in logs despite a governance rule prohibiting raw customer identifiers in captured inference data. It cannot establish the required Silvercrest University correction for: An enterprise ML team assumes a larger GPU instance always reduces cost although the model is small and memory bandwidth is not a bottleneck. The decisive acceptance check is instead: The selected instance meets the latency requirement with measured lower normalized cost than the oversized GPU option.

A synthetic slice regression is detected even though the overall average metric remains inside its broad threshold. This would verify a different operation: Create monitored business slices and evaluate approved per-segment metrics and bias indicators using retained ground-truth labels. That separate action addresses: A service tracks only average model prediction values while a narrow segment receives systematically poor results that aggregate statistics hide. It cannot establish the required Silvercrest University correction for: An enterprise ML team assumes a larger GPU instance always reduces cost although the model is small and memory bandwidth is not a bottleneck. The decisive acceptance check is instead: The selected instance meets the latency requirement with measured lower normalized cost than the oversized GPU option.

Use appropriate SageMaker Clarify post-training bias measures and valid governed outcome data with documented fairness interpretation. It does not directly address the Pinecrest Hotels case: An enterprise ML team assumes a larger GPU instance always reduces cost although the model is small and memory bandwidth is not a bottleneck. The relevant acceptance criterion for this case is: The selected instance meets the latency requirement with measured lower normalized cost than the oversized GPU option.

Add model-specific output-validity checks and application error metrics alongside endpoint health measurements. It does not directly address the Pinecrest Hotels case: An enterprise ML team assumes a larger GPU instance always reduces cost although the model is small and memory bandwidth is not a bottleneck. The relevant acceptance criterion for this case is: The selected instance meets the latency requirement with measured lower normalized cost than the oversized GPU option.

Use AWS Secrets Manager or an approved secret provider with a scoped runtime role and suppress secret values from pipeline logs. It does not directly address the Pinecrest Hotels case: An enterprise ML team assumes a larger GPU instance always reduces cost although the model is small and memory bandwidth is not a bottleneck. The relevant acceptance criterion for this case is: The selected instance meets the latency requirement with measured lower normalized cost than the oversized GPU option.

A controlled prediction from the new variant appears in the correct monitoring dataset and the old variant is not mistaken for current. This would verify a different operation: Update monitored endpoint or variant configuration and baseline references as part of the versioned deployment process. That separate action addresses: A risk model is deployed with a new model version but its SageMaker Model Monitor schedule continues capturing only the old endpoint variant. It cannot establish the required Pinecrest Hotels correction for: An enterprise ML team assumes a larger GPU instance always reduces cost although the model is small and memory bandwidth is not a bottleneck. The decisive acceptance check is instead: The selected instance meets the latency requirement with measured lower normalized cost than the oversized GPU option.

A representative request cohort produces traceable per-model score differences and unacceptable changes block promotion. This would verify a different operation: Run shadow or controlled A/B inference comparisons with logged model version and matched request IDs before full rollout. That separate action addresses: A retrained model is deployed without comparing its output distributions with the approved production model, causing unexplained score shifts. It cannot establish the required Pinecrest Hotels correction for: An enterprise ML team assumes a larger GPU instance always reduces cost although the model is small and memory bandwidth is not a bottleneck. The decisive acceptance check is instead: The selected instance meets the latency requirement with measured lower normalized cost than the oversized GPU option.

An authorized monitoring sample supports drift analysis and a security scan finds no forbidden raw identifiers in captured payloads. This would verify a different operation: Configure bounded and redacted inference data capture with access controls, sampling and retention matched to monitoring requirements. That separate action addresses: A SageMaker endpoint stores all prediction payloads in logs despite a governance rule prohibiting raw customer identifiers in captured inference data. It cannot establish the required Pinecrest Hotels correction for: An enterprise ML team assumes a larger GPU instance always reduces cost although the model is small and memory bandwidth is not a bottleneck. The decisive acceptance check is instead: The selected instance meets the latency requirement with measured lower normalized cost than the oversized GPU option.

Publish an approved data subset or governed view and grant read access only to the required machine learning project roles. It does not directly address the Oakridge Public Services case: A model's endpoint throughput suddenly falls but the operator adds instances before investigating payload size changes in incoming traffic. The relevant acceptance criterion for this case is: A controlled payload replay reproduces the bottleneck and the targeted correction restores throughput without indiscriminate scaling.

Monitor inference request schema and null-value rates against a versioned accepted contract before treating the endpoint as healthy. It does not directly address the Oakridge Public Services case: A model's endpoint throughput suddenly falls but the operator adds instances before investigating payload size changes in incoming traffic. The relevant acceptance criterion for this case is: A controlled payload replay reproduces the bottleneck and the targeted correction restores throughput without indiscriminate scaling.

Replace embedded access keys with authorized IAM roles and secure short-lived credential delivery for the notebook environment. It does not directly address the Oakridge Public Services case: A model's endpoint throughput suddenly falls but the operator adds instances before investigating payload size changes in incoming traffic. The relevant acceptance criterion for this case is: A controlled payload replay reproduces the bottleneck and the targeted correction restores throughput without indiscriminate scaling.

A controlled shifted feature batch triggers the intended drift alert while a reference batch remains within its baseline threshold. This would verify a different operation: Create a SageMaker Model Monitor data-quality baseline and schedule inference feature-distribution checks against it. That separate action addresses: A customer churn model starts receiving features with a distribution that differs sharply from training, but the endpoint tracks only CPU and availability. It cannot establish the required Oakridge Public Services correction for: A model's endpoint throughput suddenly falls but the operator adds instances before investigating payload size changes in incoming traffic. The decisive acceptance check is instead: A controlled payload replay reproduces the bottleneck and the targeted correction restores throughput without indiscriminate scaling.

The monitoring metadata links the live model package to its correct reviewed baseline and tests produce expected alerts. This would verify a different operation: Version monitoring baselines with each released model and capture a reviewed transition policy between old and new expectations. That separate action addresses: A model owner uses one unversioned monitoring baseline for every retraining run, so legitimate model improvements trigger constant false alarms. It cannot establish the required Oakridge Public Services correction for: A model's endpoint throughput suddenly falls but the operator adds instances before investigating payload size changes in incoming traffic. The decisive acceptance check is instead: A controlled payload replay reproduces the bottleneck and the targeted correction restores throughput without indiscriminate scaling.

The monitoring report labels distribution shift and actual error metrics distinctly and does not infer accuracy loss from drift alone. This would verify a different operation: Distinguish covariate drift from verified performance degradation and evaluate outcome labels and model-quality metrics separately. That separate action addresses: An operations dashboard treats a rising model input drift score as proof that prediction correctness already decreased without label evidence. It cannot establish the required Oakridge Public Services correction for: A model's endpoint throughput suddenly falls but the operator adds instances before investigating payload size changes in incoming traffic. The decisive acceptance check is instead: A controlled payload replay reproduces the bottleneck and the targeted correction restores throughput without indiscriminate scaling.