DP-600: The Fabric Analytics Engineer Exam Where Preparing Data Is Half the Test
Most Microsoft exams spread their weight evenly enough that you can afford a weak spot. DP-600 does not. Implementing Analytics Solutions Using Microsoft Fabric — the exam behind the Microsoft Certified: Fabric Data Engineer Associate credential — puts 45 to 50 percent of its weight on a single domain: Prepare data. Nearly half your score comes from getting data ready: connecting to it, transforming it, and querying it in SQL, KQL, and DAX. The other two domains split the rest evenly. That lopsided shape is the whole story of this exam. If you study it like a balanced paper, you will underprepare the half that matters most.
The exam is the analytics sibling of DP-700. Where DP-700 certifies the data engineer who builds the Fabric platform — pipelines, lakehouses, monitoring — DP-600 certifies the analytics engineer who turns what is in the lakehouse and warehouse into secured semantic models the business can actually query. Same platform, different job. DP-700 asks how data gets in and stays healthy. DP-600 asks how data becomes answers: clean, modeled, governed, and fast enough to report on. Candidates who already hold PL-300 will recognize the reporting end of the work; what is new here is everything underneath it — OneLake, the lakehouse and warehouse, and the semantic model layer that sits between raw data and the dashboard.
The exam in one paragraph
DP-600 is a single proctored exam that earns the Microsoft Certified: Fabric Analytics Engineer Associate certification. It runs 100 minutes, the passing score is 700 out of 1,000, and the fee is $165 in the United States. There are no formal prerequisites — PL-300 is the recommended on-ramp, not a requirement — and the exam is delivered through Pearson VUE, at a test center or online with a live proctor. The credential is valid for one year and renews with a free online assessment on Microsoft Learn, the same annual renewal rule as Microsoft's other associate certifications. The skills measured were updated most recently on October 19, 2026, a minor revision to the query-and-analyze objectives that left the three-domain structure and the weights unchanged. Microsoft does not publish the question count; expect roughly 40 to 60 items across multiple choice, multi-select, drag-and-drop, case studies, and performance-based tasks.
The three domains — and the one that decides your score
Prepare data (45–50%) is the domain that decides this exam. It covers connecting to data and discovering it through the OneLake catalog and the Real-Time hub, choosing between data stores — lakehouse, warehouse, eventhouse — and transforming data into analysis-ready shape: views, functions, stored procedures, star schemas, denormalizing, deduplicating, and handling nulls. Then it tests querying and analyzing that data with the Visual query editor, SQL, KQL, and DAX. The exam expects you to query in three languages, not one, and to know which language fits which job: SQL for the warehouse, KQL for event and log data, DAX for the semantic layer. The trap is depth without breadth — a candidate who is fluent in SQL but cannot read a KQL query loses marks across a domain worth half the paper. Study this domain as if it were the whole exam, because it nearly is.
Implement and manage semantic models (25–30%) is the domain where analytics engineering actually happens. It covers designing and building semantic models: storage modes, star schemas, bridge tables and many-to-many relationships, DAX variables, iterators, windowing and information functions, calculation groups, dynamic format strings, and field parameters. Then it covers optimizing them: DAX and visual performance, Direct Lake configuration, large model storage format, incremental refresh, and composite models. This is the domain PL-300 holders know best — and the one where they get overconfident. DP-600 tests the model layer harder than PL-300 does: not just whether a measure returns the right number, but whether the model is built to stay fast and correct as data grows. Direct Lake is the signature topic here. Know when it applies, what it requires of the underlying Delta tables, and what happens when it falls back.
Maintain a data analytics solution (25–30%) is the domain about keeping the solution secure, governed, and deployable after it ships. It covers access controls at every level — workspace, item, row, column, object, and file — plus sensitivity labels, endorsement, and reusable assets such as .pbit and .pbids files and shared semantic models. It also covers the lifecycle: Git version control, Power BI Desktop projects (.pbip), deployment pipelines, impact analysis, and the XMLA endpoint. Candidates treat this as the "admin" domain and study it last. It is a quarter to a third of the exam, and its questions are among the most concrete — the right answer is usually a specific control or a specific step in a pipeline, not a judgment call. Marks here are cheap if you have done the work and expensive if you have not.
The three traps that catch experienced candidates
First, studying PL-300 and calling it DP-600 preparation. PL-300 teaches you to build reports on top of a model someone else made. DP-600 tests whether you can build and govern the whole stack underneath the report. The overlap is real — semantic models, DAX, Power BI — but the exam's center of gravity is data preparation in Fabric, and PL-300 never touches OneLake, lakehouses, Dataflows Gen2, or KQL. Candidates who arrive with only PL-300 knowledge find the dominant domain unfamiliar and the maintain domain entirely new. Treat PL-300 as a prerequisite you happen to hold, not as the study plan.
Second, knowing one query language and guessing the rest. DP-600 is a trilingual exam by design: SQL, KQL, and DAX each have their own objectives, and the exam writes questions that punish single-language fluency. A KQL question about filtering event data cannot be answered by translating it into SQL in your head — the operators and the mental model are different. The same goes for DAX: a question about a calculation group or a window function is testing whether you think in DAX, not whether you can approximate it in SQL. Give each language real practice time. The cheapest marks in the dominant domain are the ones where you simply recognize what a query does.
Third, memorizing the semantic model without understanding Direct Lake and storage modes. The model domain's hardest questions are not about writing DAX — they are about choosing how the model connects to data. Import, DirectQuery, and Direct Lake each have different performance characteristics, different refresh behavior, and different failure modes, and the exam's scenarios are built around those differences. A question that describes a large Delta table with frequent updates is not asking which DAX function to use; it is asking whether Direct Lake fits and what it needs. Candidates who study measures and ignore storage modes lose the questions that separate pass from fail in this domain.
DP-600 vs. DP-700: which one first?
The two Fabric associate exams confuse candidates because they share a platform and a vocabulary. The difference is the job. DP-700 is the engineering exam: pipelines, ingestion, Spark, monitoring, capacity. DP-600 is the analytics exam: preparing data, semantic models, governance, deployment. If you build data pipelines for a living, DP-700 is your exam and DP-600 is the natural second credential — the pair together covers the full Fabric stack, and employers increasingly ask for both. If you build reports and models, DP-600 is your exam, and DP-700 is the stretch goal that teaches you what happens upstream of your model. Either order works, but study them as different exams with different centers of gravity, not as two versions of the same test. DP-700's domains are evenly weighted; DP-600's are not. That single fact should change how you split your study time.
A study plan that works
Weeks 1–2: Learn Fabric as one platform. Start with the architecture: OneLake, workspaces, lakehouses, warehouses, eventhouses, and the shared security and governance model. Build a small workspace yourself, load data into a lakehouse, and open it in a warehouse — because the exam's data-preparation questions are much easier when you have seen where data lives and how the stores differ. Read the current DP-600 skills outline — the October 19, 2026 version — and map every objective to one of the three domains. Note the shape: one dominant domain, two equal ones. Plan your weeks around that shape, not around the order the outline lists them.
Weeks 3–4: Prepare data, hands on. This is the largest share of the work, so it earns the most time. Practice getting data in through Dataflows Gen2, pipelines, and shortcuts; transform it in Spark notebooks and T-SQL; and query it in all three languages. For every pattern you build, practice the exam's question shape: a scenario with constraints, four plausible approaches, one correct fit. Keep a running list of the cues that signal which language or tool a question expects — streaming data, business-user-owned transformations, large-scale compute — because those cues are the difference between the right answer and the plausible one.
Week 5: Semantic models, deep. Build models from scratch: star schema, relationships, measures, calculation groups, field parameters. Then break them on purpose — a slow visual, a wrong total, a model that refreshes badly — and fix them, because the optimization objectives are tested through failure scenarios, not through definitions. Practice Direct Lake configuration and know its requirements and its fallback behavior cold.
Week 6: Governance, lifecycle, and timed practice. Work through the access-control levels, sensitivity labels, deployment pipelines, and Git integration as a system. Then sit full practice sets at exam pace — roughly two minutes per question across 100 minutes — and review every miss by domain. Review each wrong answer by asking whether the miss was knowledge or judgment. If knowledge, study the fact. If judgment, study scenarios: the gap is decision-making, not information. Pay special attention to misses in the prepare-data domain, since it is nearly half the exam.
Bottom line
DP-600 rewards the candidate who respects its shape: prepare data is half the test, semantic models are the craft, and governance is the quarter that is too easy to neglect. It is not a harder PL-300 and it is not a lighter DP-700 — it is the exam for the person who turns Fabric data into answers the business can trust. Study the dominant domain like it is the whole exam, learn all three query languages, and understand storage modes as well as you understand DAX. The candidates who pass are not the ones who memorized the most functions. They are the ones who can take raw data in OneLake and deliver a governed, performant semantic model on the other side.
If you want to check your readiness before paying the exam fee, the free DP-600 practice test at Study4Pass covers the same ground — exam-style questions with instant feedback, built against the current Fabric outline: https://study4pass.com/study-material/microsoft/dp-600