Rapid Battle Cards 15–28
- Conformed dimension vs local dimension — governed consistently across fact areas/marts vs specific to one use/schema.
- Bus matrix vs physical schema — process-by-dimension planning/integration roadmap vs implemented table design.
- Historical load vs ongoing update — baseline/backfill vs recurring current-state deltas.
- Timestamp CDC vs log-table CDC — select by reliable timestamps, deletes not inherent vs explicit source change records able to represent deletes.
- Log-table CDC vs transaction-log CDC — application/source-managed change table vs DBMS transaction evidence.
- Message delta vs full load — published low-latency changes vs complete-state extraction/comparison.
- Trickle feed vs messaging — source-side mini-batch accumulation vs event-bus publication/distribution.
- Messaging vs streaming — bus accumulation/subscription vs target-side queue/buffer accumulation.
- Requirements vs source inventory — business decisions/questions/KPIs first vs source catalog supporting them after scope is understood.
- Architecture vs population — end-to-end arrangement/roles/flows/controls vs executing data movement/loading.
- Data track vs technology track — meaning, sources, mappings, DQ, models/rules vs platform, storage, orchestration, security, operations.
- Mapping vs transformation — document source-target equivalence/rules vs execute intended conversion/derivation.
- Remediation vs transformation — correct defective data/root cause vs intentionally change valid data to target representation.
- Optimistic vs pessimistic loading — proceed with controlled provisional member + reconcile vs recycle/reject until dependency valid + reload.
Fast drill
- Reliable
updated_at, no delete need → timestamp CDC plausible. - Dedicated table records insert/update/delete → log-table CDC.
- DB log required to capture deletes → transaction-log CDC.
- Events accumulate on bus → messaging.
- Direct continuous events queue at target → streaming.
Source: pp. 368–379.