5.2 Last-Week Review Map

5.2 Last-Week Review Map

The final week is for consolidation, not new material. DP-700 has three domains, each weighted 30-35 percent, so points are spread evenly and you cannot afford a blind spot. Use this map to lock in the single highest-yield fact per area, then prove it with short mixed sets.

Domain 1 - Implement and manage an analytics solution (30-35%)

The spine here is the Fabric storage and governance model. OneLake is a single, logical, tenant-wide data lake built on Delta Parquet; every workspace and item stores data in it, and shortcuts virtualize external data (ADLS Gen2, S3, GCS) without copying. Access is layered: workspace roles set the coarse permission, and OneLake data access roles and item-level sharing refine it.

For application lifecycle management (ALM), remember two rules that get tested constantly: Git integration supports Azure DevOps Git and GitHub for branch-based source control, and deployment pipelines copy item definitions and metadata, not the underlying data, so a freshly deployed semantic model must be refreshed and data-source credentials stay stage-specific.

Domain 2 - Ingest and transform data (30-35%)

The most-tested distinction is which tool does which job. Pipelines orchestrate multi-step workflows (copy activity, scheduling, retries, event triggers). Dataflows Gen2 do low-code, Power Query (M) shaping for analysts. Notebooks run Spark code (PySpark, Spark SQL) for custom, distributed transformation. Eventstream handles low-code real-time ingestion and routing into a lakehouse or eventhouse, while KQL queries telemetry in an eventhouse.

Layer those tools onto the medallion architecture: bronze is raw landed data, silver is cleaned and conformed, gold is the curated, denormalized star schema ready for Power BI. Loading patterns matter: use full load for small dimensions, incremental load when a reliable change timestamp exists, and mirroring for continuous low-latency replication of a source such as Azure SQL into OneLake without custom ETL. Streaming jobs must use checkpoints, each with its own location, to resume safely after restart.

Domain 3 - Monitor and optimize an analytics solution (30-35%)

The Monitoring hub is the single place to see run status across pipelines, notebooks, dataflows, and other items - it is the answer to almost any 'where do I check the status' question. For storage performance, Delta OPTIMIZE compacts many small files into fewer large ones, and V-Order applies a read-optimized sort that speeds Power BI and SQL reads; VACUUM removes stale files past the retention window. For capacity, learn the throttling ladder driven by smoothing windows.

Last-week schedule

• Day 7-5: drill your weakest of the three domains
• Day 4-3: mixed timed sets across all three
• Day 2: error-log rules plus the OPTIMIZE/V-Order and throttling tables
• Day 1: logistics, light review, sleep

High-yield cross-domain facts worth memorizing cold

A handful of facts span the three domains and appear repeatedly. Drill these until they are reflex:

• Shortcuts vs mirroring vs copy: a shortcut points at data in place (internal OneLake or external ADLS Gen2, S3, GCS) with no copy; mirroring continuously replicates a supported source (such as Azure SQL or Cosmos DB) into OneLake; a copy activity physically moves data. Pick the one the requirement implies - 'no ETL, low latency, stays in sync' is mirroring; 'query external Delta without staging' is a shortcut.
• SQL analytics endpoint is read-only: a lakehouse exposes a read-only SQL analytics endpoint, while a warehouse supports full read-write T-SQL. If a stem needs inserts or updates via T-SQL, it needs a warehouse, not the lakehouse endpoint.
• Security layers: workspace roles are coarse; OneLake data access roles, row-level security (RLS), column-level security (CLS), and dynamic data masking (DDM) refine it. A Viewer who can list a lakehouse but cannot read its files needs a OneLake data access role; an analyst seeing masked values needs the UNMASK permission.
• Endorsement: Promoted is self-service by item owners; Certified requires an authorized reviewer.

Tool-to-language quick map

DP-700 expects fluency in SQL, PySpark, and KQL, and the language often tells you which tool the question is about.

If you can name the tool, language, and the single decisive fact for each domain without notes, the last week has done its job. Resist adding brand-new resources now - scattered review the night before a scaled exam lowers confidence more than it raises knowledge.