Exam Bundle

That observation is a valid indicator for “A pipeline reads real managed tables and the team needs to validate end-to-end behavior” when using Run an isolated integration test with controlled input and output tables. The scenario instead tests “A report consumer needs a read-only business presentation assembled from managed tables”; the relevant result is Queries return the current base-table values without a separate refresh job.

The deployed target has its expected resource names without editing shared production defaults pertains to a different control (Use a target-specific override for the job resources in the bundle) and a different need (An environment needs a controlled version of the same notebook job with separate resource names). It cannot establish that Create a view with the required projection when no persisted recomputation is needed met the stated goal.

New rows appear while previous committed rows remain present pertains to a different control (Define a streaming table over an append-only supported source) and a different need (A streaming table should preserve append-only raw events rather than reaggregate the entire source). It cannot establish that Create a view with the required projection when no persisted recomputation is needed met the stated goal.

This action solves a different issue: “Records need to be ranked within each region without collapsing individual rows.” Its normal signal is Each input record retains a rank within its region. Here the requested behavior is “Invalid rows should be counted but still retained for later audit,” so that signal is not sufficient.

Use Auto Loader `rescue` schema evolution mode with a rescued data column is appropriate when “Unexpected fields should be retained for inspection but not added to the target schema”; it would be verified by New unexpected attributes appear in the rescued data payload. It does not directly satisfy “Invalid rows should be counted but still retained for later audit,” whose relevant acceptance evidence is Expectation metrics record failures while target includes the failing rows.

Write a separate selection of failing rows to a quarantine target while the clean flow filters them is appropriate when “Bad records must be preserved in a separate quarantine dataset”; it would be verified by Quarantine contains the rejected records and silver contains only valid records. It does not directly satisfy “Invalid rows should be counted but still retained for later audit,” whose relevant acceptance evidence is Expectation metrics record failures while target includes the failing rows.

An older event does not replace a newer target value would demonstrate the separate action Add an ordering/version predicate to the matched update condition, which addresses “Only rows with a newer event version should replace an existing target record.” It does not verify the requirement “Invalid rows should be counted but still retained for later audit” following Use an expectation configured to warn without dropping rows.

That observation is a valid indicator for “Records need to be ranked within each region without collapsing individual rows” when using Use a ranking window with `PARTITION BY region`. The scenario instead tests “Invalid rows should be counted but still retained for later audit”; the relevant result is Expectation metrics record failures while target includes the failing rows.

Quarantine contains the rejected records and silver contains only valid records pertains to a different control (Write a separate selection of failing rows to a quarantine target while the clean flow filters them) and a different need (Bad records must be preserved in a separate quarantine dataset). It cannot establish that Use an expectation configured to warn without dropping rows met the stated goal.

This action solves a different issue: “Only rows with a newer event version should replace an existing target record.” Its normal signal is An older event does not replace a newer target value. Here the requested behavior is “Invalid rows should be excluded from the validated silver target without failing the entire update,” so that signal is not sufficient.

This action solves a different issue: “Bad records must be preserved in a separate quarantine dataset.” Its normal signal is Quarantine contains the rejected records and silver contains only valid records. Here the requested behavior is “Invalid rows should be excluded from the validated silver target without failing the entire update,” so that signal is not sufficient.

The alternative Pin the dependency to a validated version and redeploy has a distinct purpose—An external dependency silently changes behavior after an unpinned upgrade. It would not establish the required outcome (Invalid rows should be excluded from the validated silver target without failing the entire update); the necessary verification concerns The target omits invalid rows and expectation counters record drops.

This signal belongs to the adjacent case “Bad records must be preserved in a separate quarantine dataset,” whose remedy is Write a separate selection of failing rows to a quarantine target while the clean flow filters them. The observed result for the present requirement is The target omits invalid rows and expectation counters record drops.

A second update adds new source rows without replacing prior accepted rows would demonstrate the separate action Read the incremental source into a streaming table instead of overwriting a static table, which addresses “A target needs append-only ingestion rather than destructive daily replacement.” It does not verify the requirement “Invalid rows should be excluded from the validated silver target without failing the entire update” following Configure an expectation with DROP behavior.

That observation is a valid indicator for “Records need to be ranked within each region without collapsing individual rows” when using Use a ranking window with `PARTITION BY region`. The scenario instead tests “Invalid rows should be excluded from the validated silver target without failing the entire update”; the relevant result is The target omits invalid rows and expectation counters record drops.

This action solves a different issue: “Bad records must be preserved in a separate quarantine dataset.” Its normal signal is Quarantine contains the rejected records and silver contains only valid records. Here the requested behavior is “Any violation of a strict financial invariant must fail the pipeline update,” so that signal is not sufficient.

The alternative Use a fail-on-new-columns schema evolution strategy has a distinct purpose—A strict fixed schema must reject unexpected incoming fields until a reviewed change. It would not establish the required outcome (Any violation of a strict financial invariant must fail the pipeline update); the necessary verification concerns An invalid test row aborts the update with expectation failure details.

This action solves a different issue: “A task exceeds a strict 15-minute processing limit and should fail fast.” Its normal signal is The job marks an overlong task timed out and follows its failure path. Here the requested behavior is “Any violation of a strict financial invariant must fail the pipeline update,” so that signal is not sufficient.

The job marks an overlong task timed out and follows its failure path would demonstrate the separate action Set a supported task timeout matching the SLA, which addresses “A task exceeds a strict 15-minute processing limit and should fail fast.” It does not verify the requirement “Any violation of a strict financial invariant must fail the pipeline update” following Configure an expectation with FAIL behavior.

That observation is a valid indicator for “Bad records must be preserved in a separate quarantine dataset” when using Write a separate selection of failing rows to a quarantine target while the clean flow filters them. The scenario instead tests “Any violation of a strict financial invariant must fail the pipeline update”; the relevant result is An invalid test row aborts the update with expectation failure details.

This signal belongs to the adjacent case “An analytical transformation is supported on serverless pipelines with no custom cluster management need,” whose remedy is Select supported serverless Lakeflow pipeline execution. The observed result for the present requirement is An invalid test row aborts the update with expectation failure details.

This action solves a different issue: “A streaming Delta sink must accept new append rows and preserve committed offsets.” Its normal signal is New records accumulate exactly once at supported sink boundaries after restart. Here the requested behavior is “Bad records must be preserved in a separate quarantine dataset,” so that signal is not sufficient.

Configure an expectation with DROP behavior belongs to the separate case “Invalid rows should be excluded from the validated silver target without failing the entire update,” where The target omits invalid rows and expectation counters record drops would be expected. That differs from the condition in this stem: “Bad records must be preserved in a separate quarantine dataset.”

Apply a full-record DISTINCT projection to the intended columns is appropriate when “A raw events dataset must remove exact duplicate records without altering other fields”; it would be verified by Identical duplicate rows collapse to one while distinct rows stay. It does not directly satisfy “Bad records must be preserved in a separate quarantine dataset,” whose relevant acceptance evidence is Quarantine contains the rejected records and silver contains only valid records.

This signal belongs to the adjacent case “A Python pipeline definition should register a managed materialized view with declarative SQL,” whose remedy is Use the supported pipelines Python decorator for a materialized view function. The observed result for the present requirement is Quarantine contains the rejected records and silver contains only valid records.

This signal belongs to the adjacent case “An existing expectation is too broad and flags legitimate nulls in an optional attribute,” whose remedy is Narrow the expectation predicate to the required mandatory fields. The observed result for the present requirement is Quarantine contains the rejected records and silver contains only valid records.

MERGE commits without multiple source rows matching the same target row pertains to a different control (Deduplicate to one deterministic source row per target key before MERGE) and a different need (A source batch contains duplicate rows matching one target key and MERGE becomes ambiguous). It cannot establish that Write a separate selection of failing rows to a quarantine target while the clean flow filters them met the stated goal.

Configure appropriate Auto Loader schema evolution and persisted schema location is appropriate when “New optional JSON fields should be captured without discarding previously known columns”; it would be verified by A restart after a new column update recognizes the evolved schema. It does not directly satisfy “An existing expectation is too broad and flags legitimate nulls in an optional attribute,” whose relevant acceptance evidence is Valid optional-null rows pass while null mandatory fields are flagged.

Plan a new checkpoint and controlled state reinitialization or migration is appropriate when “A data engineer changes stateful grouping keys and the existing checkpoint is incompatible”; it would be verified by The restarted query initializes the intended state schema without checkpoint errors. It does not directly satisfy “An existing expectation is too broad and flags legitimate nulls in an optional attribute,” whose relevant acceptance evidence is Valid optional-null rows pass while null mandatory fields are flagged.

This action solves a different issue: “Bad records must be preserved in a separate quarantine dataset.” Its normal signal is Quarantine contains the rejected records and silver contains only valid records. Here the requested behavior is “An existing expectation is too broad and flags legitimate nulls in an optional attribute,” so that signal is not sufficient.

This signal belongs to the adjacent case “A raw events dataset must remove exact duplicate records without altering other fields,” whose remedy is Apply a full-record DISTINCT projection to the intended columns. The observed result for the present requirement is Valid optional-null rows pass while null mandatory fields are flagged.

That observation is a valid indicator for “A full snapshot source is authoritative and the target should delete keys absent from it” when using Use a carefully scoped `WHEN NOT MATCHED BY SOURCE THEN DELETE` clause. The scenario instead tests “An existing expectation is too broad and flags legitimate nulls in an optional attribute”; the relevant result is Valid optional-null rows pass while null mandatory fields are flagged.

Quarantine contains the rejected records and silver contains only valid records pertains to a different control (Write a separate selection of failing rows to a quarantine target while the clean flow filters them) and a different need (Bad records must be preserved in a separate quarantine dataset). It cannot establish that Narrow the expectation predicate to the required mandatory fields met the stated goal.