Data Science and Digital Solutions: Core Concepts and Exam Traps
What this domain covers
This domain tests whether you can reason across the full data lifecycle rather than recite tool names. You must understand data types and structures (structured, semi-structured, unstructured; tabular vs. log vs. text), data quality dimensions (accuracy, completeness, consistency, timeliness, validity, uniqueness), and the difference between descriptive, diagnostic, predictive, and prescriptive analytics. Expect questions on data governance: stewardship roles, data lineage, metadata, master data management, and privacy obligations such as consent, minimization, and retention limits.
On the digital-solutions side, know the solution stack: data sources and ingestion (batch vs. streaming), storage (warehouse vs. data lake vs. lakehouse), transformation and orchestration, the semantic/metrics layer, visualization, and operational integration via APIs. Understand cloud service models (IaaS, PaaS, SaaS) and shared-responsibility boundaries, plus basic security controls: authentication vs. authorization, role-based access, encryption at rest and in transit, and audit logging.
Traps that appear in questions
- Correlation presented as causation, or a predictive model described as proving why something happened.
- Accuracy vs. precision confused with statistical precision; also validity (conforms to rules) vs. accuracy (matches reality).
- Data lake vs. warehouse framed as a maturity ladder rather than a design choice; lakes do not enforce schema on write by default.
- Oversampling or dropping nulls offered as harmless fixes; both change the population being analyzed.
- Dashboards treated as governed sources when the metric layer, not the chart, is authoritative.
- Anonymization claims that ignore re-identification risk from combining quasi-identifiers.
Rule of thumb
Before accepting any data answer, ask three questions: What is the grain of a row? Where did the number come from (lineage)? Who is allowed to see it? If a proposed fix changes the population, flag it as a governance decision, not a cleaning step.
Worked example
A dashboard shows sales up 12% after a filter change. The grain shifted from order-line to order-header, so multi-line orders collapsed. Correct answer: reconcile at the documented grain before interpreting the trend.