Confusion Matrix, Changed-Fact Drills & Cram Route
High-risk confusion matrix
| Pair |
Deciding distinction |
Changed-fact clue |
| Validity vs Accuracy |
formal rule/domain vs reality |
formatted email belongs to wrong person → valid form, inaccurate |
| Timeliness vs Currency |
delivery delay vs currentness |
on-time three-day-old balances → timely delivery, poor currency |
| Completeness vs Integrity |
required presence vs valid relationship |
populated orphan FK → integrity |
| Consistency vs Uniqueness |
agreement vs one-per-entity |
duplicate customer → uniqueness |
| Profiling vs DQ Rule |
observed statistic vs approved expectation |
12% nulls vs ≥99% must be populated |
| Initial Assessment vs Monitoring |
baseline discovery vs repeated control |
first investigation vs daily scorecard |
| Correction vs Root Cause |
repair record vs remove recurrence cause |
fix rows vs change UI/process |
| DQ SLA vs Scorecard |
service commitment vs summarized condition |
resolve within 4 hours vs completeness 97% |
| Parse vs Standardize vs Enrich |
decompose vs change representation vs add |
split name vs normalize vs add geocode |
| DQ Function vs Project |
continuing capability vs temporary initiative |
project ends; function persists |
Twelve changed-fact drills
- Email quality: 98% populated, requirement 95% → completeness passes. If 40% bounce → accuracy becomes larger problem.
- Supplier file: arrives at 8 a.m. but contains last week’s customers → timeliness may pass; currency fails. Current values arriving noon → timeliness fails.
- Customer master: two IDs, same identity → uniqueness. Single record with another customer’s address → accuracy.
- Order records: CUSTOMER_ID populated but parent missing → integrity. CUSTOMER_ID null → completeness.
- Profile finding: 7% STATE null → potential issue only. Approved optional rule may make nulls valid.
- Issue response: 50k rows corrected monthly → correction, cause likely remains. Unchangeable upstream with cheaper controlled correction can justify ongoing correction.
- Incident workflow: severe issue unassigned → assignment/escalation failure. Ownership clear but obsolete rule → rule management.
- SLA: quality threshold exists but notification/deadline absent → incomplete SLA. Commitments exist but no measurement → measure/monitor gap.
- Parsing: free-form name must be decomposed → parsing. Components known but need enterprise format → standardization.
- Correction confidence: proposed fixes with confidence + Steward review → manually-directed. Deterministic auto-commit → fully automated.
- Readiness: leaders support DQ but actual state unknown → build objective state first. State/risk/priorities known → broader program more feasible.
- Governance: reports show failures but business units cannot agree priority/ownership → Governance. Ownership agreed but processing defect remains → process/system remediation.
Decision-clue route
- Identify threatened business use/consumer requirement.
- Decide whether stem is about criticality, dimension, rule, measurement, issue, or improvement/control.
- Classify exact dimension failure before selecting a tool or fix.
- Determine baseline assessment vs ongoing control.
- Separate symptom correction from root-cause prevention.
- Identify accountable owner/decision mechanism.
- Select tool/technique/artifact last.
10-minute cram route
- 0–2: fit for purpose + nine dimensions.
- 2–4: eight principles + criticality.
- 4–6: PDCA + seven activities.
- 6–8: rule management, monitoring, issues, SLA, response.
- 8–10: profiling/RCA, prevention/correction, tool distinctions, Governance links.
← Targets 18–34 · Visual Atlas →