Scenario Lab — 16–22
16 — Change Made Without History
A logical entity changes for a new requirement; team overwrites the file with no explanation.
Best: version/change control recording Why, What/How, When, Who, Where, with appropriate analyst/architect/system-owner review.
Weaker: rely on email memory.
Changed fact: a temporary personal layout-only sketch changes but the governed semantic model does not → formal model-change approval may not be triggered; once governed content changes, control applies.
17 — Model Looks Neat but Misses Requirements
Beautiful diagram follows naming rules but omits two required business facts.
Best: Scorecard capture requirements and likely completeness fail first.
Weaker: polish layout/naming further.
Changed fact: all requirements are present but keys/relationships cannot produce a valid implementable design → structural soundness becomes primary.
18 — Star or Snowflake?
Campus dimension contains City, State, Region; team proposes separate normalized hierarchy tables.
Best: movement toward snowflake because dimension hierarchy is normalized into component tables.
Trade-off: dimension normalization vs collapsed star simplicity/navigation.
Weaker: call it normalization of the fact table; the change is in the dimension.
Changed fact: keep City/State/Region as columns in one Campus dimension directly attached to fact → star-like.
19 — Fresh Every Time or Precomputed?
Expensive report must return instantly each morning; overnight refresh is acceptable.
Best: materialized view — stored/instantiated result refreshed at predetermined time.
Trade-off: some freshness/storage/refresh management for retrieval speed.
Changed fact: latest committed values must always be reflected at query time and performance is acceptable → standard view may be better.
20 — Split the Wide Table
180-column table; most screens use 15 columns, rare large fields seldom read; row count manageable.
Best: vertical partitioning because the problem is column separation.
Other direction: horizontal partitioning divides rows.
Weaker: denormalization; the requirement is not duplicate/combine but split a wide structure.
Changed fact: billions of rows must be separated by region/date while retaining same columns → horizontal partitioning.
21 — Vendor Industry Model
University buys education-industry reference model; leadership wants to use it exactly and skip requirements workshops.
Best: industry model is a useful starting/reference point but must be customized and validated against local requirements, terminology, rules, and scope.
Weaker: either use it blindly or reject it entirely.
Changed fact: team uses it only as a comparison/reference and validates every adopted structure → source-aligned.
22 — Scorecard Says 92, Database Disagrees
Logical model scores well on readability/naming/definitions/structure, but production tables changed directly and model no longer matches stored data.
Best: Scorecard Metadata matches actual data is failing. Governance failure also includes maintenance + version/change control. Modelers/architects and implementation teams must reconcile reality and the governed model.
Weaker: merely reformat the diagram.
Changed fact: model and stored data agree, but two required business facts are absent from both → requirements capture/completeness becomes primary.