CDMP cheat sheet

DAMA-DMBOK wheel: 11 knowledge areas grouped into Data Governance (center), Data Architecture, Data Modeling & Design, Data Storage & Operations, Data Security, Data Integration & Interoperability, Document & Content Management, Reference & Master Data, Data Warehousing & BI, Metadata, Data Quality.

Data lifecycle phases: Plan, Design & Enable, Build/Produce, Use & Maintain, Enhance, Dispose/Archive; each phase has data-quality and metadata checkpoints.

Data vs. information: data are raw recorded facts; information is data processed into context and meaning; knowledge is information integrated with experience.

ACID properties for transactional databases: Atomicity (all-or-nothing), Consistency (valid state transitions), Isolation (concurrent transactions do not interfere), Durability (committed changes survive failure).

CAP theorem: a distributed data system can simultaneously guarantee at most two of Consistency, Availability, Partition tolerance; partition tolerance is mandatory in practice, so the real trade-off is C vs. A.

Data storage types: relational (row-oriented), columnar (analytics), key-value, document, graph, and object stores; choose by access pattern, consistency need, and schema volatility.

Data security controls: authentication, authorization (role-based or attribute-based), encryption at rest and in transit, masking, tokenization, and audit logging.

Data management maturity is assessed across capability levels (e.g., 0–5) per knowledge area; maturity models measure process capability, not data quality directly.

Three model levels: conceptual (business entities and relationships, technology-neutral), logical (attributes, keys, normalization, no DBMS specifics), physical (tables, indexes, partitions, data types for a specific DBMS).

Entity-relationship notation: entity = noun; relationship = verb; cardinality options are 1:1, 1:M, M:N; optionality is mandatory or optional on each end.

Normalization: 1NF removes repeating groups; 2NF removes partial dependencies on part of a composite key; 3NF removes transitive dependencies on non-key attributes; BCNF removes every determinant that is not a candidate key.

Denormalization deliberately introduces redundancy (pre-joined tables, derived columns) to reduce join cost in read-heavy or analytical workloads; it trades write complexity and anomaly risk for query speed.

Dimensional modeling: fact tables hold measures and foreign keys; dimension tables hold descriptive context; star schema has denormalized dimensions; snowflake schema normalizes dimension hierarchies.

Slowly changing dimensions: Type 1 overwrites (no history), Type 2 adds a new row with version/effective dates (full history), Type 3 adds a previous-value column (limited history).

Surrogate keys are system-generated, stable, and independent of business keys; natural keys are business-meaningful and may change, so facts should reference surrogates.

Grain declares what one fact-table row represents; it must be defined before choosing dimensions and measures, and every measure must be additive at that grain.

Metadata types: business (definitions, owners, rules), technical (schemas, data types, ETL mappings, lineage), operational (job logs, row counts, refresh times), and administrative (policies, access rights).

Metadata is managed through a metadata repository; a data catalog is the user-facing interface over that repository, supporting search, lineage, and glossary terms.

Data lineage records the path of data from source to target, including transformations; column-level lineage is required for impact analysis and root-cause tracing.

Impact analysis uses lineage in reverse (what downstream objects change if this column changes); root-cause analysis uses lineage forward (where did this bad value originate).

A business glossary defines business terms, synonyms, owners, and rules; it is the authoritative vocabulary linked to technical metadata.

Metadata integration patterns: a centralized repository (single store), a distributed registry (federated pointers), or a hybrid with a central index and local stores.

Metadata quality is assessed by completeness, accuracy, currency, and consistency; stale metadata is a common cause of failed data projects.

Metadata drives data quality rules, security classification, retention schedules, and self-service analytics; without it, data users cannot find or trust data.

Data quality dimensions: completeness, accuracy, consistency, timeliness, uniqueness, validity, and conformity; each dimension needs an operational definition and measurement method.

Completeness = non-null (or non-missing) values / total values; accuracy compares data to a verified source of truth; consistency checks agreement across systems or within a record.

Timeliness compares data availability against the required latency; uniqueness detects duplicate records; validity checks conformance to format, range, or domain rules.

Data profiling measures column statistics (null rate, distinct count, min/max, pattern frequency) to discover issues before defining rules.

Data quality rule types: completeness, validity, range, pattern, referential integrity, uniqueness, and cross-field consistency rules.

Data quality management cycle: define dimensions and rules, profile, measure, remediate root causes, monitor, and report; remediation targets process causes, not just bad records.

Data cleansing techniques: standardization, parsing, matching/deduplication, enrichment, and correction; cleansing without root-cause fix causes recurring defects.

Data quality metrics should be reported as trends with thresholds and ownership; a single score without dimension breakdown hides actionable problems.

Data governance is the exercise of authority and control over data management, covering policy, standards, decision rights, and accountability; it is not the same as data management execution.

Operating model components: data governance council (strategic decisions), data stewardship (tactical execution), data owners (accountable for a domain), and data custodians (technical care).

Decision rights define who decides on data definitions, access, quality thresholds, and issue resolution; unclear decision rights are a primary governance failure.

A data policy states principles and mandatory rules; standards specify measurable requirements; procedures describe how to comply; guidelines are recommended practices.

Data stewardship roles: business data steward (definitions, quality rules) and technical data steward (implementation, lineage); stewards are usually part-time domain experts.

Governance metrics include policy compliance rate, issue resolution time, glossary coverage, and stewardship participation; metrics must be tied to business outcomes.

Data governance maturity progresses from ad hoc to defined, managed, measured, and optimized; each level requires documented processes and assigned accountability.

Regulatory drivers (privacy, retention, reporting) require governance to map data to legal obligations, define retention/disposal rules, and evidence compliance.

A data warehouse is a subject-oriented, integrated, time-variant, nonvolatile collection of data supporting management decisions; it is built from operational sources via ETL/ELT.

Inmon approach: enterprise data warehouse as a normalized central store, then data marts; Kimball approach: dimensional bus matrix and conformed dimensions across marts.

Data mart types: dependent (sourced from the warehouse), independent (sourced directly from operational systems), and hybrid; independent marts risk inconsistent definitions.

Conformed dimensions are shared, identically defined dimensions across fact tables, enabling drill-across and consistent reporting.

ETL vs. ELT: ETL transforms before loading into the warehouse; ELT loads raw data first and transforms inside the target platform, common with cloud warehouses.

BI delivery layers: operational reports, dashboards, OLAP cubes, ad hoc query, and advanced analytics; each has different latency and aggregation requirements.

OLAP operations: slice, dice, drill-down, roll-up, and pivot; these navigate a multidimensional cube of dimensions and measures.

Data warehouse architecture layers: staging, integration, presentation, and metadata/access layers; staging isolates source extraction from transformation and load.

Reference data are controlled code sets used to classify or categorize other data (country codes, currency codes, status codes); they are relatively static and shared across systems.

Master data are the core business entities (customer, product, employee, supplier, location) shared across processes; master data management creates a single trusted version.

MDM implementation styles: registry (index and links, no central golden record), consolidation (central read-only copy), coexistence (central copy with bidirectional sync), and centralized (authoritative master store).

A golden record is the single trusted version of a master entity, built by survivorship rules that choose or merge attribute values from source systems.

Entity resolution (matching) uses deterministic rules (exact keys) and probabilistic matching (similarity scores) to identify duplicates; thresholds control false positives and false negatives.

Reference data management requires a governed code-set owner, versioning, effective dates, and controlled distribution to consuming systems.

MDM data domains need defined hierarchies (e.g., customer household, product category) and relationships; hierarchy management is part of the MDM platform scope.

MDM success metrics: match rate, duplicate reduction, golden-record completeness, cross-system consistency, and time to onboard a new source.

Data integration patterns: extract-transform-load (ETL), extract-load-transform (ELT), change data capture (CDC), replication, federation, and virtualization.

Batch integration moves data on a schedule; real-time integration uses streaming or messaging; near-real-time uses micro-batch; choose by latency requirement and source capability.

Change data capture identifies inserts, updates, and deletes since the last run using timestamps, log-based capture, or triggers; log-based CDC has the least source impact.

Data virtualization queries distributed sources through a logical layer without physically moving data; it suits volatile or latency-sensitive access but can strain source systems.

Interoperability requires agreed data formats, semantics, and interfaces; standards (e.g., XML, JSON, industry message formats) reduce point-to-point mapping effort.

Integration design must handle data quality checks, error handling, restart/recovery, idempotency, and audit logging; silent failures are a common integration defect.

A hub-and-spoke integration architecture centralizes transformation and reduces point-to-point interfaces; a bus or messaging backbone decouples producers and consumers.

Data integration governance includes interface contracts, schema versioning, SLA monitoring, and lineage capture from source to target.