Deep Battle Cards A–F
A — Validity vs Accuracy
Definition: Validity = conformance to domain/type/range/format/precision/allowed values. Accuracy = correspondence with reality.
Purpose: allowed vs true.
Owners/users: Stewards define domains; process/system owners implement controls; trusted sources/consumers help verify accuracy.
Inputs: definitions/domains/reference lists vs verified source/reality evidence.
Outputs: validation exceptions vs accuracy-verification findings.
Deciding distinction: rule conformance vs reality.
Scenario A: email matches format but belongs to another customer → valid form, inaccurate.
Scenario B: AGE=247 violates approved range → validity.
Trap: “wrong data” does not automatically mean accuracy; identify what makes it wrong.
Memory: VALID = allowed; ACCURATE = real.
Source: pp. 430–432.
B — Completeness vs Integrity
Definition: Completeness asks whether required data exists. Integrity asks whether values/relationships are coherently linked.
Scope: presence vs relationship/referential coherence.
Inputs: mandatory/conditional requirements vs parent-child/reference rules.
Outputs: missing values/records vs orphan/broken-relationship exceptions.
Deciding distinction: presence vs relationship.
Scenario A: CUSTOMER_ID is null → completeness.
Scenario B: CUSTOMER_ID='123' but customer 123 does not exist → integrity.
Trap: a populated FK can be complete but still violate integrity.
Memory: Complete = there; Integrity = connected.
Source: pp. 430–431.
C — Timeliness vs Currency
Definition: Timeliness = delay from capture/update to availability. Currency = how current the represented fact remains.
Scope: service/processing latency vs age/volatility.
Inputs: capture/availability timestamps vs last-update time and volatility expectation.
Outputs: latency metrics vs stale/current indicators.
Deciding distinction: when I get it vs how fresh it is.
Scenario A: today’s market data arrives after a 5-minute SLA → timeliness.
Scenario B: address last verified eight years ago in volatile population → currency.
Trap: “old” is not automatically timeliness.
Source: pp. 431–432.
D — Consistency vs Uniqueness
Definition: Consistency = expected agreement across representations/sets/time. Uniqueness = one representation per real entity.
Scope: cross-value/system agreement vs identity/deduplication.
Inputs: equivalent fields/sets vs identifying attributes/keys.
Outputs: mismatch exceptions vs duplicate candidates.
Deciding distinction: agreement vs one-per-entity.
Scenario A: two IDs represent same customer → uniqueness.
Scenario B: source and warehouse balance differ when they should match → consistency.
Trap: not every cross-system mismatch is uniqueness.
Source: pp. 430–432.
E — Profiling vs Initial Assessment vs Monitoring
Definition: Profiling is a technique that reveals patterns. Initial assessment is a baseline management activity. Monitoring is ongoing repeated measurement.
Purpose: discover → judge baseline → sustain control.
Inputs: data sets, rules, criticality, thresholds.
Outputs: statistics/patterns → assessed findings/priorities → recurring measures/trends/alerts.
Deciding distinction: technique vs lifecycle stage.
Scenario A: first null-distribution exploration on new scope → profiling supporting initial assessment.
Scenario B: hourly invalid-record score and threshold alert → monitoring.
Trap: the same profiling tool can support both assessment and monitoring; tool ≠ management activity.
Memory: PROFILE sees; ASSESS judges baseline; MONITOR watches.
Source: pp. 443–447, 451.
F — Prevention vs Correction vs Root-Cause Remediation
Definition: Prevention stops future defects; correction repairs existing data; RCA/remediation removes underlying factors causing recurrence.
Owners: process/system owners own prevention/root-cause changes; DQ facilitates; Stewards may approve correction.
Outputs: controls/process changes; corrected records; permanent remediation plan.
Deciding distinction: future defect vs existing defect vs cause.
Scenario A: add UI validation → prevention.
Scenario B: repair 50k invalid records → correction.
Trap: repeated cleansing is not a substitute for changing the process that produces the defects.
Memory: PREVENT before; CORRECT after; ROOT CAUSE stops repeat.
Source: pp. 428, 444–445, 457–459.