Survivorship Rules

The most common survivorship strategy assigns a trust score to each source system. When a conflict arises, the value from the highest-ranked source wins.

• CRM > ERP: A customer's phone number from the CRM might survive over the billing system's record.
• Certified > Inferred: A manually verified address overrides one generated by a fuzzy matching algorithm.
• Implementation: A static lookup table maps source_id to an integer rank; the merge engine selects MAX(rank).

This rule assumes the most recent update is the most accurate. The value with the latest modified_at timestamp survives the merge.

• Use Case: Ideal for volatile attributes like job titles, stock quantities, or current location.
• Granularity: Can be applied at the attribute level (e.g., phone changes frequently) rather than the whole record.
• Risk: A system clock error or a batch backfill can incorrectly crown a stale value as the winner.

When data completeness is the priority, survivorship logic selects the most information-rich value.

• Longest String: Chooses the value with the highest CHAR_LENGTH. A full middle name survives over an initial.
• Least Nulls: A sparse record from System A is rejected in favor of a fully populated record from System B.
• Composite Density: A scoring function counts non-null fields; the record with the highest completeness score becomes the survivor.

Also known as majority rules, this strategy selects the value that appears most often across all duplicate records.

• Mechanism: If 4 out of 5 source records say NY and 1 says New York, NY survives.
• Tie-Breaker: In a deadlock, a secondary rule (like recency or priority) is invoked.
• Statistical Validity: Effective for crowdsourced or sensor data where individual errors are random, but the consensus is likely correct.

Rarely does a single rule govern an entire entity. A survivorship matrix defines distinct logic per attribute.

• Name: Longest value rule to preserve full legal names.
• Phone Number: Recency rule to capture the latest mobile device.
• Annual Revenue: Source priority rule (Finance system over Sales forecast).
• Address: USPS-validated value always survives over unverified strings.

A non-negotiable requirement for data governance. The merge engine must log why a value survived and which values were deprecated.

• Metadata: The golden record stores survivor_source, survivor_rule, and deprecated_values as JSON.
• Lineage: Enables rollback if a survivorship rule is later found to be flawed.
• Compliance: Essential for GDPR 'right to rectification' and financial auditing standards.