Troubleshooting & Problem Resolution Flashcards

A machine learning model's accuracy degrades steadily in production though the code is unchanged. What is the most likely cause?

Answer: Data or concept drift in the input data

Drift in input distributions or relationships erodes model performance over time without code changes.

Which capability most helps trace a wrong value in a report back to its originating system?

Answer: Data lineage

Data lineage maps how data flows and transforms from source to report, enabling root cause tracing.

A critical data incident affects regulatory reporting due tomorrow. What should the CDO prioritize first?

Answer: Containing impact and communicating with stakeholders and compliance

Incident response starts with containment and timely stakeholder communication before long-term fixes.

Duplicate customer records keep appearing after each CRM integration run. Which solution addresses the root cause?

Answer: Implementing master data management with matching and survivorship rules

MDM matching and survivorship rules consolidate records systematically at integration time.

Pareto analysis applied to data quality issues helps the team to do what?

Answer: Focus on the few root causes driving most defects

The 80/20 principle shows that a small set of causes typically creates most problems.

A nightly ETL job intermittently fails with timeouts only at month-end. What is the best first diagnostic step?

Answer: Correlate failures with data volumes and resource utilization at month-end

Pattern-based failures should be correlated with volume and resource metrics to isolate capacity issues.

What is the main purpose of a blameless post-incident review for a data outage?

Answer: To document lessons learned and systemic fixes without fear of punishment

Blameless reviews encourage honest analysis so systemic weaknesses can be fixed.