Lesson 8 — Governance, Data Model Scorecard, and Decision Rules
A data model is not complete because a modeling session ended. Chapter 5 treats models as governed assets that must be reviewed, controlled, and maintained.
Three governance control areas
1. Modeling and design standards
Standards are the rulebook: expected notation, naming, modeling practices, required Metadata, and related conventions.
2. Design review
A design review is a quality checkpoint for a specific model. Chapter 5 emphasizes correctness, completeness, and consistency.
A multidisciplinary review may involve modelers, analysts, business SMEs/stewards, architects, designers, DBAs, and system owners as appropriate.
If a review finds unresolved conflicts, the model is reworked rather than approved merely because a meeting occurred. When a final unresolved design decision remains, the owner of the system represented has the source-defined final authority.
3. Version / change control
Version control preserves change lineage. Record:
- Why the change happened;
- What / How changed;
- When;
- Who;
- Where in the model/scope.
Memory hook: rulebook vs checkpoint vs history.
Review vs maintenance
Review tests quality at a point in time. Maintenance keeps the model synchronized as requirements, processes, and implementations change.
If teams alter physical tables directly and never update the logical/physical model, the organization no longer has trustworthy model Metadata.
Data Model Scorecard — 100 points
The current Mastery Lab preserves these ten categories and weights:
| Category | Points |
|---|---|
| Capture requirements | 15 |
| Completeness | 15 |
| Match the declared modeling scheme/level | 10 |
| Structural soundness | 15 |
| Appropriate use of generic structures | 10 |
| Naming standards | 5 |
| Readability | 5 |
| Definitions | 10 |
| Consistency with enterprise models | 5 |
| Metadata matches actual data | 10 |
| Total | 100 |
A polished diagram can fail badly if requirements are missing. A complete-looking model can fail structural soundness. A model can score well on readability and naming but still fail because the actual database drifted away from the governed Metadata.
Scorecard distinction examples
- Two required business facts are absent → capture requirements / completeness.
- Model claims relational but violates the selected scheme’s structural rules → scheme fit.
- Relationships/keys cannot produce a valid implementation → structural soundness.
- Names are cryptic/inconsistent → naming.
- Diagram is hard to read → readability.
- Model says one thing but production stores another → Metadata matches actual data.
Twenty high-value deciding rules
- Data Architecture vs Modeling & Design: enterprise blueprint/alignment vs precise model/design work.
- Entity vs instance: type vs occurrence.
- Relationship vs foreign key: business association vs relational implementation mechanism.
- Cardinality vs arity: number of occurrences vs number of entity types.
- Attribute vs domain: property vs valid values.
- Simple/composite/compound/surrogate: one attr / 2+ / 2+ FKs / generated meaningless ID.
- Super/candidate/primary/alternate: any unique / minimal unique / chosen / unchosen candidate.
- Identifying vs non-identifying: migrated parent key enters child PK vs remains non-PK FK.
- Relational vs dimensional: operational business rules vs analytic navigation.
- Fact vs dimension: measurement vs descriptive context.
- Grain vs cardinality: fact-row meaning vs relationship participation.
- Star vs snowflake: collapsed vs normalized dimension structures.
- CDM vs LDM vs PDM: high-level business / detailed tech-independent / technology-specific.
- Normalization vs denormalization: remove redundancy/dependency defects vs deliberately add/combine for justified physical need.
- Normalization vs abstraction: dependency cleanup vs supertype/subtype inheritance.
- Forward vs reverse: requirements downward vs existing database upward.
- View vs materialized view: virtual on-demand vs stored/instantiated result.
- Vertical vs horizontal partition: columns vs rows.
- Pattern vs industry model: generic reusable structure vs broad industry reference starting point.
- Tool vs modeler: software stores/transforms representations; human judgment establishes business meaning and appropriate design.
Final teach-it-back
You should be able to explain Chapter 5 as one flow:
Discover requirements → agree meaning and scope → model entities/relationships/attributes/domains and keys → select an appropriate scheme → refine through CDM/LDM/PDM → normalize and abstract logically → adapt the physical design deliberately → preserve lineage and reusable Metadata → review against standards and Scorecard criteria → version and maintain the model as reality changes.