Free ADA-C02 practice test: 30 questions with answers

A SELECT with GROUP BY and a date filter returns the aggregate without modifying data. UPDATE, DELETE, and COPY INTO change data or load files and are not needed for a read-only analysis.

Clustering keys improve micro-partition pruning for frequently filtered columns on large tables. A comment, a view key, and renaming do not reduce scanned data in Snowflake.

Time Travel lets users query or restore historical table states within the retention period. Cloning a different database, Snowpark procedures, and dashboard refreshes do not restore the table.

A zero-copy clone shares underlying storage until data changes, so it is cheap to create. CSV exports, copy-paste, and laptop files are slow and duplicate data.

A masking policy enforces role-based masking at query time. Comments and manual copies rely on people and risk leaks, and sharing a password breaks access control.

Row access policies filter rows by user context. Comments, per-user databases, and warehouse limits do not restrict which rows a user sees.

Granting only SELECT through a role follows least privilege. ACCOUNTADMIN and OWNERSHIP give far more rights, and shared accounts remove accountability.

Masking, row access policies, and least privilege protect sensitive data. Shared admin logins, personal exports, and disabled query history increase exposure or remove accountability.

A clearly labelled axis, ideally starting at zero for bar charts, avoids exaggerating changes. Hiding labels and adding 3D effects reduce accuracy.

Interactive filters let stakeholders explore without SQL and keep results live. Screenshots, copies per region, and emailed raw data are static or cumbersome.

Grouping minor categories and sorting by value makes the chart readable. More categories, unique colours for everything, and removed labels reduce readability.

A finding-led title gives the message immediately, with detail below. Source, full tables, and version history bury the message.

Reconciling row counts and control totals locates where the difference arose. Cosmetic changes and deleting reports do not diagnose the cause.

Constraints or validation checks prevent or flag nulls early. Ignoring nulls, silently replacing them, and removing columns hide the problem.

A defined key and deduplication in the transformation fix the cause for all consumers. Dividing counts, hiding rows in one chart, and manual daily clean-up are partial or fragile fixes.

Row counts and control totals confirm completeness. Chart colours, dashboard counts, and table name length do not show whether data is complete.

Auto-suspend and right-sizing reduce idle and oversized compute. Maximum size all the time, many new warehouses, and disabled logging increase cost or reduce insight.

The query profile shows scan volumes, spills, and time per step, which point to bottlenecks. Titles, chart counts, and colours do not affect runtime.

Resource monitors set credit quotas and notify before limits are reached. Reminders, annual reviews, and shutting down all warehouses do not control spending in a timely way.

Resource monitors and right-sized warehouses with auto-suspend control spend. Frequent refreshes, largest warehouses for everyone, and disabled result reuse increase cost.

A two-sample test assesses whether the difference is likely real, after checking assumptions. Comparing averages alone, picking the larger one, and counting customers do not assess chance.

The median is robust to extreme values and represents a typical value better than the mean in skewed data. The maximum, sum, and count do not describe typical values.

Correlation alone does not show that one metric causes the other, so further investigation is needed. The other statements draw unsupported conclusions.

Representative samples and reported uncertainty support sound conclusions. Stopping on favourable results, selective outlier removal, and reusing data for hypothesis selection bias the result.

A shared semantic definition ensures consistent metrics everywhere. Separate definitions, the oldest one by default, and renaming keep the inconsistency.

Cortex Analyst uses a semantic model to answer natural-language questions over governed data. A PDF, printed names, and file folders do not interpret questions.

Defining the relationship on the key tells the model how the tables join. Colours, constants, and passwords do not describe the data structure.

Secure data shares give read-only access to selected objects without copying data. CSV emails create copies, admin access is excessive, and public pages expose data.

A secure view with row filtering controls what each consumer sees and protects the logic. Sharing unfiltered tables, PDFs, and manual copies break the control.

Consumers read shares without copies, and providers can share selected objects. Shared data is read-only for consumers, exporting is not required, and consumers cannot change the provider's source.