Question 350
Production records start having missing customer_age fields that were nearly always populated during training. What monitoring check is appropriate?
Explanation
Correct Answer
Null-rate changes represent data-quality degradation and may require a different response from ordinary statistical distribution drift.
Incorrect Answers
Choose an observation window with enough representative data for the monitored metric. This is useful for the distinct issue: A monitor uses a one-hour window with only a handful of samples and produces unstable drift readings. The relevant technical rationale or evidence for that issue is: Small windows can make distribution comparisons unreliable; sufficient samples improve interpretation of detected change. Here the reported issue is: Production records start having missing customer_age fields that were nearly always populated during training. The relevant response is: Monitor data-quality signals such as null rates for critical features. The first result cannot verify this different correction: Null-rate changes represent data-quality degradation and may require a different response from ordinary statistical distribution drift.
Configure alert notifications and an escalation owner for threshold breaches. It addresses another case: A drift monitor detects a large shift but the platform team has not configured any alert recipient. For that case, the distinguishing justification or acceptance check is: Monitoring results must reach accountable operators before they can trigger investigation or approved retraining. The present case instead states: Production records start having missing customer_age fields that were nearly always populated during training. It calls for Monitor data-quality signals such as null rates for critical features. Showing success for the other case would not establish: Null-rate changes represent data-quality degradation and may require a different response from ordinary statistical distribution drift.
The alternative, Segment production drift results by relevant region or customer cohort, belongs to a different technical condition: A drift dashboard reports a statistic but no indication which new population is affected. There the source-specific mechanism or proof is: Overall drift values can hide localized data changes, so cohort analysis improves diagnosis and remediation targeting. The condition asked about here is: Production records start having missing customer_age fields that were nearly always populated during training. A suitable correction is Monitor data-quality signals such as null rates for critical features; the targeted justification or result is: Null-rate changes represent data-quality degradation and may require a different response from ordinary statistical distribution drift.