Name the three Medallion layers and their one‑phrase purpose.
Bronze (raw landing), Silver (clean/validated), Gold (curated/business‑ready).
Which Delta feature lets you query older snapshots?
Time Travel (VERSION/ TIMESTAMP AS OF).
What format name enables Auto Loader’s incremental discovery?
cloudFiles (e.g., .format("cloudFiles")).
What’s the UC object path order?
catalog.schema.table.
One benefit that Lakehouse provides over separate lake + warehouse.
Unified storage/compute with Delta for ACID + governance + one copy for BI & AI.
Command to compact small files and optionally cluster by columns.
OPTIMIZE (with optional ZORDER BY).
One key difference between Auto Loader and COPY INTO.
Auto Loader = incremental/stream‑friendly with checkpointing; COPY INTO = idempotent batch copy.
Minimum grants to read a table?
USAGE on catalog & schema and SELECT on the table
In UC, what’s the difference between a Table and a Volume?
Table = managed/external Delta with schema & SQL access; Volume = governed file storage (no table schema).
What Delta statement is most often used for upsert/CDC into a target?
MERGE INTO with matched/unmatched clauses.
In streaming ingestion, what’s the purpose of a checkpoint?
Tracks progress/offsets & state for exactly‑once semantics and recovery.
Name a scalable way to give analysts read access to all Silver tables.
Grant at schema level to a group/role (or use shares), not per‑table.
With frequent new fields from streaming JSON, which layer/feature absorbs change best?
Bronze with Auto Loader schema evolution (and optionally DLT expectations up‑stream).
Why can over‑aggressive VACUUM be dangerous?
It deletes old files, breaking Time Travel/streaming readers; risk of data loss if retention too low.
Name a common strategy to dedupe late‑arriving records
Watermark + window + row_number() over keys (keep rn=1), or Delta constraints + MERGE on latest.
Common approach to row‑level filtering in UC.
Dynamic views referencing identity (e.g., current_user()/identity()) joined to entitlements table.