5.2 High-Yield Traps & Confusions

Where DP-600 Sets Its Traps

Microsoft builds plausible wrong answers from features that sound interchangeable but are not. The exam rarely tests an obscure fact in isolation; it tests whether you can pick the right tool when two or three close cousins are all listed as options. If you can state the precise difference between each pair in one sentence, both standalone multiple-choice items and case-study questions become dramatically easier.

The pattern to watch: a question describes a need, lists four Fabric features, and three of them technically exist but solve a slightly different problem. Your job is to match the need to the one feature whose purpose is an exact fit — for example, a need for near-real-time KQL telemetry maps to an eventhouse, not a lakehouse, even though both can store the data. Memorize the deciding rule, not a vague sense that the features are 'similar.'

There are two reliable tells in a distractor-heavy question. First, a plausible-but-mismatched option names a real feature that does the wrong job — for example, offering incremental refresh for a Direct Lake freshness need. Second, an overscoped option proposes a heavier mechanism than required — offering a deployment pipeline when the task is merely to flag content as trusted. Train yourself to ask, for every option, "what problem is this feature actually built to solve?" and discard any whose purpose does not match the stated need.

The High-Yield Confusion Table

Storage Mode Matched to Refresh Technique

A recurring trap deliberately pairs the wrong maintenance technique with a storage mode. Keep these three facts crisp:

• Incremental refresh partitions and refreshes Import tables. It is not how Direct Lake stays current.
• Direct Lake freshness is governed by framing (also called reframe): when the underlying Delta files change, the model reframes to the new data — no scheduled import.
• A DirectQuery model has nothing to import-refresh; it queries the source live every time.

So when a scenario says "the Direct Lake model must reflect new data automatically" and an option proposes configuring incremental refresh on the Direct Lake model, that option is almost always the distractor. The correct answer relies on framing, or on ensuring the source Delta tables are written/optimized. Worked example: a nightly ETL writes to a Delta table feeding a Direct Lake model — no refresh schedule is needed; the model frames to the latest committed version on the next query.

One more nuance the exam likes: when a Direct Lake model receives a query it cannot answer in memory (for instance, the data exceeds the capacity guardrails for the SKU), it does not fail — it falls back to DirectQuery against the SQL analytics endpoint. So "the report still returns results but feels slower after a large load" describes fallback behavior, not a broken refresh. Distinguish three phrases cleanly: framing (Direct Lake freshness), fallback (Direct Lake serving a query via DirectQuery), and refresh (Import only).

Endorsement Is Not Deployment; Classification Is Not Security

Two governance confusions appear in question after question.

Endorsement (Promoted/Certified) signals trust in content. It does not move content between environments — deployment pipelines (Development → Test → Production) do that. So "promote the dataset to production" is solved by a deployment pipeline, not by clicking Promoted.

A sensitivity label (e.g., Confidential, Highly Confidential) classifies and can encrypt an item for compliance and downstream protection (the label persists into exported files). It does not decide who can open an item — workspace roles (Admin/Member/Contributor/Viewer) and item-level permissions do.

Hold two phrases in mind under pressure: trust signal vs promotion mechanism, and classification vs access control. Sorting each governance question into the correct bucket clears a whole cluster of distractors. A label on a model does not stop a Viewer from reading it; a Viewer role or a denied permission does.

A related pairing the exam tests is Dataflow Gen2 vs pipeline in an orchestration question. If the need is "clean and reshape data with Power Query steps," that is a Dataflow Gen2. If the need is "run these activities in sequence, on a schedule, with a copy step and a notebook," that is a pipeline — which can itself invoke a dataflow. They are complementary, so a question that frames them as either/or when the real answer is "a pipeline that runs the dataflow" is fishing for the candidate who has only used one of them.