DMLI cheat sheet

Data management is the development and execution of architectures, policies, practices, and procedures that manage the full data lifecycle.

The DAMA-DMBOK wheel organizes knowledge areas around data governance at the center; governance is the hub that coordinates all other disciplines.

The data lifecycle phases are: plan, design and enable, create and obtain, store and maintain, use and enhance, share and reuse, archive, and purge.

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

CAP theorem: a distributed system can simultaneously guarantee only two of consistency, availability, and partition tolerance.

Metadata types: business metadata (definitions, rules), technical metadata (schemas, data types), operational metadata (job logs, row counts).

A data steward is accountable for a data domain's quality and definitions; a data owner holds decision rights and budget authority for that domain.

Master data describes core business entities (customer, product, employee); reference data is a controlled set of permissible values (codes, statuses).

Governance operating model components: decision rights, accountability structures, policies and standards, and enforcement mechanisms.

GDPR lawful bases for processing include consent, contract, legal obligation, vital interests, public task, and legitimate interests; at least one must apply.

Data subject rights under GDPR include access, rectification, erasure, restriction, portability, and objection; access requests must generally be answered within one month.

A Data Processing Agreement (DPA) is required between a controller and a processor and must specify subject matter, duration, nature and purpose, and security measures.

Data classification tiers commonly run Public, Internal, Confidential, Restricted; controls (encryption, access, retention) are assigned per tier.

A Record of Processing Activities (ROPA) documents purpose, categories of data and subjects, recipients, retention, and security measures for each processing activity.

Data retention schedules define minimum and maximum retention periods; legal hold suspends destruction for data relevant to litigation or investigation.

Segregation of duties separates authorization, custody, and reconciliation of data to prevent unilateral error or fraud.

Six core data quality dimensions: completeness, accuracy, consistency, timeliness, uniqueness, and validity.

Completeness is measured as non-null records divided by total records; validity is measured as records passing defined rules divided by total records.

Data profiling statistics include null counts, distinct counts, min/max, frequency distributions, and pattern detection.

ETL order: extract, transform, load into a target store; ELT loads raw data first and transforms inside the target platform.

Idempotency means rerunning a pipeline produces the same result; it is required for safe retries and backfills.

Data lineage traces data from source to target, supporting impact analysis (what breaks downstream) and root-cause analysis (where a defect originated).

Slowly changing dimensions: Type 1 overwrites history, Type 2 adds a new row with effective dates and a current flag, Type 3 adds a prior-value column.

Reconciliation controls compare source and target record counts and control totals to detect lost or duplicated records during integration.

Three levels of data model: conceptual (business entities and relationships), logical (attributes, keys, normalized), physical (tables, indexes, storage).

Normalization: 1NF removes repeating groups, 2NF removes partial dependencies on part of a composite key, 3NF removes transitive dependencies on non-key attributes.

Dimensional modeling uses fact tables (measures, foreign keys) and dimension tables (descriptive context); star schema has denormalized dimensions, snowflake schema normalizes them.

Grain is the level of detail of a fact table row and must be declared before choosing measures or dimensions.

A slowly changing dimension is the standard technique for preserving dimension history; surrogate keys decouple warehouse keys from source business keys.

Kimball approach builds subject-oriented data marts conformed by shared dimensions; Inmon approach builds a normalized enterprise warehouse first.

Data Vault modeling uses hubs (business keys), links (relationships), and satellites (descriptive attributes with history).

Architecture patterns include data lake (schema-on-read), data warehouse (schema-on-write), data lakehouse (warehouse semantics over lake storage), and data fabric (virtualized integration layer).

Analytics maturity ladder: descriptive (what happened), diagnostic (why), predictive (what will happen), prescriptive (what should be done).

BI delivery modes: operational reports (scheduled, fixed), analytical dashboards (interactive), self-service (governed ad hoc), and embedded analytics.

A KPI requires an explicit definition, formula, data source, owner, and target; an undefined KPI is not measurable.

Leading indicators predict future outcomes; lagging indicators confirm outcomes already realized.

Measures are aggregations over facts (sum, count, average); dimensions are the axes used to slice measures.

Additive measures can be summed across all dimensions, semi-additive across some (balances over time), and non-additive across none (ratios, percentages).

Dashboard design rule: one primary question per dashboard, most important metric in the top-left reading position, and no more than 5-9 visual elements.

Statistical significance does not imply practical significance; correlation does not establish causation without a controlled design.

Ethical data use principles: transparency, purpose limitation, data minimization, fairness, accountability, and respect for individual autonomy.

A conflict of interest exists when a personal interest could improperly influence professional judgment; the standard response is disclosure and recusal.

Informed consent requires disclosure of purpose, data involved, and consequences, plus a voluntary and revocable affirmative act.

Bias sources in analytics include sampling bias, survivorship bias, measurement bias, and historical bias embedded in training labels.

De-identification reduces re-identification risk but does not eliminate it; k-anonymity requires each record to be indistinguishable from at least k-1 others on quasi-identifiers.

Change management adoption depends on sponsorship, stakeholder involvement, clear communication of benefit, and training; technology alone rarely drives adoption.

A RACI matrix assigns Responsible, Accountable, Consulted, and Informed roles; exactly one role should be Accountable per deliverable.

Professional codes (e.g., data ethics frameworks) require balancing stakeholder interests, documenting decisions, and escalating unresolved ethical concerns through defined channels.