Which Address Wins? The Survivorship Rule

Opening Scene

The clerk has already confirmed it: the two file cards really are the same citizen. Now comes the part that actually matters. One card lists an address from three years ago; the other, filed eight months back by the utilities office, lists a newer one. One card has a phone number; the other field is blank. One card spells the surname “Meier”; the other spells it “Meyer,” and the citizen’s own signature on a recent form settles which is correct. The clerk doesn’t average the two addresses or leave both fields on the merged file. She applies a rule — the most recent verified entry wins for address, a non-blank value beats a blank one for phone, and the citizen’s own signature outranks any office’s typed transcription for spelling. That rule, applied consistently and recorded for future audit, is what survivorship actually is.

In Plain English

Survivorship rules are the deliberate, documented policies that determine which field value wins when two or more matched source records disagree. Confirming that two records describe the same entity, as Article 3 covered, only gets you halfway to a golden record — you still need a principled way to decide, field by field, which of the disagreeing values actually gets written into the merged record. Common survivorship strategies include most-recent-wins, most-trusted-source-wins (a fixed hierarchy ranking which system is authoritative for a given field), most-complete-wins (preferring a populated value over a blank one), and most-frequently-seen-wins (when the same value appears independently across several sources). A mature MDM program rarely uses just one strategy for every field — it typically applies a different rule per field, matched to how that field actually behaves.

The Old Way

Before survivorship is treated as a discipline in its own right, organizations tend to default to one of a few blunt instincts:

• “Newest wins,” applied to everything — a reasonable default for something like a phone number, which people update when it changes, but a poor choice for something like a legal entity name, where an old, verified value is often more trustworthy than a recent, unverified one.
• “The important system wins,” applied to everything — treating the ERP or CRM as authoritative across the board, even for fields that a different system genuinely tracks better, the same mistake Article 1 identified in the context of blind primary-system trust.
• No rule at all, decided case by case — someone simply picks whichever value looks right at the moment of merging, a decision that isn’t documented, isn’t repeatable, and can’t be explained the next time the same conflict recurs on a different record.

The discipline survivorship rules bring isn’t complexity for its own sake — it’s making a decision once, deliberately, instead of remaking an inconsistent version of the same decision every single time a conflict appears.

What’s Changing (and Why AI Is the Reason)

1. AI-assisted survivorship suggestions can recommend a likely-correct value per field based on patterns like source reliability history, recency, and cross-field consistency, rather than requiring a steward to apply a single fixed rule blindly to every case. This moves survivorship from a rigid, one-size-fits-all policy toward a more nuanced recommendation that still respects an underlying rule but can flag exceptions worth a second look.
2. AI models can learn which sources have historically proven more reliable for specific fields, refining a static source-hierarchy rule into a data-informed, continuously updated one. A source that used to be reliable for addresses but has recently started producing more corrections can be down-weighted automatically, rather than waiting for a steward to notice the pattern manually.
3. AI agents consuming golden records benefit enormously from an auditable survivorship trail, since a value’s provenance — which source it came from, and why it won — becomes essential context when an agent needs to explain or justify a decision it made using that data. Without that trail, even a correctly resolved golden record is a black box to any system trying to reason about how much to trust it.

The Metaphor, Fully Extended

Registry Element	Master Data Management Concept
Two confirmed-matching file cards with a different address on each	Two matched source records with a conflicting field value
The rule “the most recent verified entry wins for address”	A most-recent-wins survivorship rule, applied to a specific field
The citizen’s own signature outranking a typed transcription for spelling	A most-trusted-source-wins rule, ranking sources by reliability per field
A non-blank phone number beating an empty field	A most-complete-wins survivorship rule
A registry assistant flagging that one district office’s entries have needed frequent correction lately	AI-assisted, data-informed refinement of a static source-reliability hierarchy

Quick Recap

• Survivorship rules are the deliberate, documented policies that decide which field value wins when matched source records disagree.
• Common strategies include most-recent-wins, most-trusted-source-wins, most-complete-wins, and most-frequently-seen-wins, typically applied differently per field rather than uniformly.
• The default instincts organizations fall into without formal survivorship rules — blanket recency, blanket source trust, or ad hoc case-by-case decisions — all fail to scale or stay auditable.
• AI-assisted survivorship suggestions and data-informed source-reliability scoring refine, but don’t replace, the underlying governance judgment behind each rule.

Where This Fits in the Series

Articles 1 through 4 covered the foundational building blocks: what a golden record is, why copies diverge, how matching confirms two records are the same entity, and how survivorship rules resolve their conflicts. Article 5 moves into core technique, contrasting deterministic and probabilistic approaches to matching itself.