Trace why a task did not run by walking the scheduler's dependency edges to whatever blocked it, and read sensor waits as first-class tasks that can time out.

Decide what deserves a page: separate the transient failures a retry absorbed from exhausted retries and SLA misses, then build the on-call triage report unaided.

Write a batch load that owns exactly one date partition, driven by a run parameter instead of the wall clock, and prove it survives a rerun.

Apply a redelivered change batch to a dimension when two changes for one key carry the same timestamp, and keep the merge idempotent.

Keep the write path a blind append and make the read path idempotent instead, then choose between all three idempotency patterns without guidance.

Re-run the incremental loop you already know against a control table keyed by pipeline, and name the pattern in the two cloud vocabularies interviewers use.

Measure exactly how many rows a strict watermark load dropped without raising an error, then repair the pipeline with a rolling reprocessing window that self-heals.

Reprocess a date range from a backfill manifest, then rebuild the downstream aggregate for exactly the affected weeks so the mart stops lying.

Build the whole daily pipeline in one script, stage then load then audit, and rehearse the spoken design answer against a written model answer.

Materialize a staging-to-mart model chain in the order its ref graph dictates, and derive that order from the graph itself with a recursive CTE.

Collapse the four generic tests into one violation list, then build a model against a hand-written fixture the way a unit test does.

Answer what breaks downstream when a source changes, separate dev builds from prod builds, then ship the level capstone unaided.