CDMP

DAMA CDMP Data Management Fundamentals Exam: free practice test

Start with 10 free questions to see what needs more practice. Finish to see your score, weakest units and every option explained.

Back to DAMA CDMP Data Management Fundamentals Exam

Practice test questions with answers and explanations

Try the questions in the test above first, then open each one here to see the correct answer and why every other option is wrong.

01An agency holds sensitive health records that law bars it from releasing, yet its policy says its data should be FAIR. Which action best honours the policy within this limit?

Practice question Unit: Data Governance

  1. APublish rich metadata and clear rules for requesting access, while the records stay restricted ✓ Correct answer

    The principles separate data from metadata, so detailed metadata and clear access rules give a high degree of FAIRness even when the data itself stays restricted.

  2. BPublish only a short title for each record set, since detailed descriptions could expose sensitive content

    A bare title is not rich metadata; discovery and reuse need detailed descriptions, which can be written without revealing record content.

  3. CSet the policy aside for these records, because data that cannot be public cannot be made findable

    Because the principles are modular and treat metadata separately, restricted data can still be made findable and its access route made clear.

  4. DRelease a fully open copy of the records under a permissive licence so that others can find and reuse them

    An open public copy would breach the legal bar on release, and the principles do not require sensitive data to be published openly.

Reference: The FAIR Guiding Principles for scientific data management and stewardship (Scientific Data, 2016), §The FAIR Guiding Principles in detail

02In a Lambda architecture, a dashboard shows the last hour of sensor totals from the speed layer. The next morning the batch layer finishes processing that same hour, and its total differs by 2% from what the dashboard showed. Which total should analysts treat as the better figure for that hour, and why?

Practice question Unit: Data Architecture

  1. AThe speed total, because it was calculated closest to the time the events occurred and so reflects them more faithfully

    The speed layer favours low latency over accuracy; being computed sooner does not make it more faithful, and its figures are later replaced by batch results.

  2. BThe average of the two totals, because each layer is expected to hold part of the true value and errors then offset

    The layers are not complementary halves of a value; the batch result is a full, more accurate computation, so blending it with the hot-path estimate adds error.

  3. CNeither total, because a mismatch shows the raw events were overwritten and both views need rebuilding

    A small difference is expected between the paths, and the raw data at the batch layer is immutable and append-only, so nothing is overwritten.

  4. DThe batch total, because the cold path computes accurately across the full data and supersedes the hot path result ✓ Correct answer

    The speed layer trades accuracy for low latency, and its results are later updated with the more accurate cold path output, so the batch total is the better figure.

Reference: Big data architectures (Azure Architecture Center, Microsoft Learn), §Lambda architecture

03In a class model, Vehicle has a property trailers with multiplicity 0..*. Bicycle is a subclass of Vehicle and must never have any trailers, while Vehicle and its other subclasses stay unchanged. Which modeling step fits the UML rules?

Practice question Unit: Data Modeling and Design

  1. AMark trailers in Bicycle as derived so its values are computed

    Derived means a value may be computed from other information, so a derived property can still produce values rather than none.

  2. BRedefine trailers in Bicycle with a multiplicity of 0..0 ✓ Correct answer

    A zero-to-zero multiplicity is the UML way to constrain a more general classifier's property so that the specializing class holds no values.

  3. CSet trailers in Bicycle to 0..1 to limit it to a single value

    An upper bound of 1 still allows one trailer, so a Bicycle could hold a value, violating the rule that it never has any.

  4. DMark trailers in Bicycle as read-only after initialisation

    Read-only blocks updates only after an initial value is assigned; it does not forbid the property from holding values.

Reference: OMG Unified Modeling Language (UML) Version 2.5.1, §7.5.3 MultiplicityElement semantics, PDF p.76; read-only and derived semantics, PDF p.150 and p.154

04Two concurrent Repeatable Read transactions each sum a different group of rows and insert the total into the other group, then both commit. The combined result matches no one-at-a-time execution order. Which fix addresses this in PostgreSQL?

Practice question Unit: Data Storage and Operations

  1. ARun both at Read Committed so each statement sees the latest commits

    Read Committed gives weaker guarantees, so statements can interleave even more freely; it cannot make the outcome match any serial order.

  2. BKeep Repeatable Read and rely on its guarantee of a stable view

    A stable snapshot blocks dirty, nonrepeatable and phantom reads here, yet it does not detect dependency cycles between what concurrent transactions read and write.

  3. CKeep Repeatable Read and retry any transaction that raises a concurrent-update error

    The transactions modify different rows, so no concurrent-update error is raised; the retry logic never triggers and both commits succeed.

  4. DRun both at Serializable and retry whichever transaction is rolled back ✓ Correct answer

    Serializable tracks read/write dependencies and aborts one of the pair; the retried transaction then produces a result consistent with some serial order.

Reference: PostgreSQL 18 Documentation, 13.2 Transaction Isolation (PostgreSQL Global Development Group, 2026), §13.2.2 Repeatable Read Isolation Level and §13.2.3 Serializable Isolation Level

05A public agency keeps two lists with the same fields (name, address, phone) and the same number of people, on the same server with the same staff access, and no law or contract sets special duties for either. One list is newsletter subscribers; the other is applicants for a disability benefit. Which factor most justifies a higher PII confidentiality impact level for the second list?

Practice question Unit: Data Security

  1. AThe legal obligation to protect the confidentiality of the data

    The scenario states that no law or contract adds special duties, so this factor does not separate the two lists.

  2. BThe number of individuals who can be identified from the list

    Both lists identify the same number of people, so quantity cannot explain a difference in rating.

  3. CThe purpose for which the personal data is collected and used ✓ Correct answer

    The same fields can warrant different levels depending on why they are held; disclosure for a benefit application can harm people more than disclosure of a newsletter list.

  4. DThe sensitivity of the individual data fields it holds

    The fields are identical in both lists, so field sensitivity gives the same result for each.

Reference: NIST SP 800-122, Guide to Protecting the Confidentiality of Personally Identifiable Information (PII) (NIST, April 2010), §3.2.4 Context of Use, PDF p.20

06A nightly pipeline overlaps extraction, transformation and loading, with several parallel workers loading the same target table. Workers now block each other on writes, and rerunning a failed worker creates duplicate rows. Which redesign keeps the overlap while fixing both problems?

Practice question Unit: Data Integration and Interoperability

  1. AMake loading single-threaded while extraction and transformation continue to run in parallel

    One loader removes contention but gives up parallel loading, and a rerun of a failed load can still insert the same rows twice.

  2. BSplit the work along partition boundaries such as date or tenant so each worker owns a slice ✓ Correct answer

    Disjoint partitions remove write contention, and a failed slice can be rerun against only its own data, so retries are safe.

  3. CAdd more parallel workers so that a failed slice is quickly picked up by several others

    More workers on the same table increase contention, and several workers redoing one slice multiplies the duplicate rows.

  4. DAdd a deduplication step at the end of the run so reruns cannot leave duplicates in reports

    Cleaning duplicates afterwards hides one symptom but leaves the workers contending for the same table, so write blocking continues.

Reference: Extract, transform, load (ETL) (Microsoft Learn, Azure Architecture Center), section 'Extract, transform, load (ETL) process'

07A manufacturer's ERP and product lifecycle system hold the same finished goods under different IDs and descriptions, which has led to excess inventory. The two systems also code the colour red as "R" and "Red". Which pairing of remedies fits these two problems?

Practice question Unit: Master and Reference Data Management

  1. AAdd a surrogate key to every product record, and leave the colour values as they are in each system

    A surrogate key identifies a row but does not tell that two rows describe the same product, and unreconciled colour values still block combined reporting.

  2. BRetire the colour codes, and merge product records only when their descriptions are identical

    Descriptions differ between the systems, so an identical-description rule would find few duplicates, and dropping the colour codes loses the categorisation.

  3. CMatch and merge the duplicate product records into one master, and map the colour values with a crosswalk ✓ Correct answer

    Duplicate products are one entity held twice, so they need a merged master; differing colour codes are one list coded two ways, which a crosswalk reconciles.

  4. DMap the product IDs with a crosswalk and keep both records, and match and merge the colour values into one

    A crosswalk alone leaves both duplicate product records alive, so excess inventory persists, and colour values are mapped between code sets rather than merged as records.

Reference: Microsoft Purview and partner MDM tool Master Data Management (MDM) (Microsoft Learn), MDM use cases (Product data; Reference data)

08An analyst needs sales totals by year, by year and month, by year, month and day, and a grand total, and nothing else. Which grouping operator suits this best?

Practice question Unit: Data Warehousing and Business Intelligence

  1. AGROUP BY alone, because it already returns every level of subtotal and the grand total

    Plain GROUP BY gives only the finest level; subtotals need ROLLUP or CUBE, or separate unioned queries.

  2. BROLLUP, because it returns only the hierarchical super-aggregates of the listed columns ✓ Correct answer

    ROLLUP yields just the progressively coarser totals plus the grand total, which matches the linear year-month-day request exactly.

  3. CCUBE, because date columns that are derived from one another make a rollup meaningless

    The reasoning is reversed: because the columns depend on one another, a full cube would be meaningless and ROLLUP is the better fit.

  4. DCUBE, because it adds the hierarchical subtotals together with every other combination

    CUBE does include those subtotals but also produces meaningless combinations, so it returns more than the request asks for.

Reference: Data Cube: A Relational Aggregation Operator (Gray et al., Microsoft Research, 1997), §3 CUBE and ROLLUP Operators, PDF p.9

09A distribution has no access URL written on it, but it is linked to a data service whose endpoint URL is https://api.example.org/stops. A consumer applies the DCAT axioms to this description. What does it conclude?

Practice question Unit: Metadata Management

  1. AThe endpoint address is also inferred as the distribution's access URL ✓ Correct answer

    DCAT axiomatizes access URL as the chain from access service to endpoint URL, so the endpoint address is inferred as the access URL.

  2. BThe endpoint address is the landing page of the distribution's dataset

    A landing page is a property of the dataset, and nothing in the chain from service to endpoint assigns it to the dataset.

  3. CNothing, since an access URL must always be written on the distribution

    The axiom is stated precisely so that the access URL need not be repeated by hand when a service link and its endpoint are given.

  4. DThe endpoint address is the download URL, a direct link to the file

    Download URL is reserved for direct links to a downloadable file, whereas an API endpoint is a service address that gives access through queries.

Reference: Data Catalog Vocabulary (DCAT) - Version 3 (W3C Recommendation, 22 August 2024), §6.8.9 Property: access URL; §6.8.10 Property: access service

10A UK school accepts applications by date of birth. A US parent enters 09/08/2016 meaning 8 September, but the system reads it as 9 August, and a child below the age limit is admitted. What best describes the data?

Practice question Unit: Data Quality

  1. AAccuracy fails: the date is well formed yet records a birth date that is not the real one ✓ Correct answer

    The value looks like a normal date, so it passes format checks, yet it misrepresents the true birth date and fails accuracy.

  2. BValidity fails: the date breaks the expected format and should have been rejected on entry

    Wrong meaning behind a well-formed value is exactly what format and range checks cannot catch.

  3. CConsistency fails: the school register and the application hold two different date formats

    Only one source holds the date here, so there is no second format to contradict the stored value.

  4. DTimeliness fails: the application reached the school after the intake deadline had passed

    No timing problem is described; the harm comes from a misread date, not from a late entry.

Reference: The Government Data Quality Framework (UK Government Data Quality Hub, gov.uk web page), §Data quality dimensions – how to measure your data quality / Accuracy

CDMP

Full prep for this exam

All of this in one purchase.

  • Full mock exam, same as the real one100 questions in 90 minutes. You get a total score and a score for every unit.
  • Practice unit by unit632 questions, every option explained with its source. Starts with your weakest units.
  • Complete Review Guide 2026–2027 (PDF)Ordered by each unit's weight in the exam. With lifetime access only.

Choose your access

One step before payment

Continue with your Google account and you go straight to the payment page.

Check my score freeFull prep