Entity Resolution Policy¶
Procedure — instantiates Entity Individuation Criteria Design
A standing procedure for deciding whether two records refer to the same entity by applying an explicit same-as criterion rather than raw token matching.
Two records that look alike are not thereby the same entity, and two that look different are not thereby distinct — yet most systems decide the question with whatever string comparison is closest to hand. Entity Resolution Policy is the operational procedure that replaces that reflex with a governed same-as decision: given two or more presentations (records, accounts, observations), it applies an explicit identity criterion to rule same, distinct, or undetermined, and then binds the surviving identifier accordingly. Its defining move is the one the archetype insists on — the same-as decision must rest on a stated warrant, not collapse into a matching score or a shared key. A high similarity score is evidence that feeds the rule; it is never permitted to be the rule. The policy is the running machinery that turns a criterion into thousands of consistent, auditable merge-or-not calls.
Example¶
A retail bank is consolidating customer records across its mortgage, deposit, and credit-card systems. Two records share a name and date of birth but differ in address and tax ID. Naïve matching would either merge them (creating a false composite customer who now appears to breach a lending limit) or split a genuine person into two (fragmenting their relationship and mis-pricing their risk). The Entity Resolution Policy handles this by criterion, not by resemblance: for a natural person, the same-as warrant requires a verified government identifier match or two independent corroborating attributes plus no disqualifying conflict. Here the tax IDs differ and no override applies, so the policy rules distinct — and, because the case sits near the threshold, routes it to a review queue rather than auto-merging.
When the true match is confirmed later — the second tax ID was a data-entry transposition — the policy merges the records under a single customer identifier, records the losing identifiers as aliases pointing to the survivor, and preserves the lineage so the merge can be undone if it proves wrong. The identifier now follows the identity decision instead of standing in for it.
How it works¶
- Assemble candidate sets. Block and compare presentations to find plausible same-entity groups, using similarity only to narrow the field — never to decide it.
- Apply the same-as warrant. Run the explicit identity criterion (required matches, disqualifying conflicts, evidence thresholds) to classify each pair as same, distinct, or undetermined.
- Preserve uncertainty. Where evidence does not justify a hard call, hold the case open and route it for review rather than forcing a merge or split.
- Bind identifiers to the decision. On a confirmed merge, designate a surviving identifier, retire the others as aliases pointing to it, and keep reversible lineage so a wrong merge can be unwound.
Tuning parameters¶
- Evidence threshold — how strong the same-as warrant must be before auto-merging. Higher thresholds cut false merges but leave more true matches unresolved.
- Error asymmetry — whether false merges or false splits are treated as the worse error. In fraud and lending, a false merge is usually costlier and the policy leans conservative.
- Auto vs. review band — how wide the "undetermined" zone routed to humans is. A wider band is safer but more labor.
- Reversibility depth — how much lineage is retained to unwind a bad merge. More depth costs storage but makes the procedure safely correctable.
When it helps, and when it misleads¶
Its strength is consistency at scale: it makes the same-as decision the same way every time, on a stated warrant, with the identifier bound downstream of that warrant — exactly the discipline that prevents the archetype's signature failure of mistaking a key for the entity.
Its failure mode is that it is only as sound as its identity criterion, and it is tempting to let a machine-learning match score quietly become the criterion — the classic slide from evidence to warrant. Statistical record linkage from the Fellegi–Sunter model onward gives principled match probabilities, but a probability is an input to the rule, not a license to skip it.[1] Aggressive merging also produces plausible, hard-to-detect false composites; conservative splitting fragments real entities. The guarding discipline is to keep the warrant explicit and separable from any score, preserve uncertainty instead of forcing borderline calls, and retain reversible lineage so the inevitable wrong merges can be found and undone.
How it implements the components¶
identity_criterion— it operationalizes the same-as rule: the explicit warrant, applied uniformly, that decides when two presentations are one entity.identifier_binding_layer— on each ruling it binds identifiers to the decision, retiring losers as aliases pointing to the surviving key with reversible lineage.
It does not decide the unity_criterion (part-whole composition) — that is drafted in the Entity Definition Workshop; this policy resolves same-as between whole presentations, not what composes one whole.
Related¶
- Instantiates: Entity Individuation Criteria Design — the running procedure that turns the same-as criterion into consistent merge-or-not decisions.
- Consumes: Individuation Criteria Charter — supplies the identity criterion the policy applies.
- Sibling mechanisms: Entity Definition Workshop · Identity and Unity Test Checklist · Individuation Criteria Charter · Master Entity Registry · Split/Merge Decision Tree · Count Impact Assessment · Edge-Case Adjudication Panel · Versioned Identity Rulebook
Editorial Notes¶
Form Classification¶
Form family: Protocol, Workflow & Routine
Rationale: Entity Resolution Policy operates as a repeatable ordered procedure or handoff sequence that coordinates action because it a standing procedure for deciding whether two records refer to the same entity by applying an explicit same-as criterion rather than raw token matching.
Independent corroboration: The frozen evidence defines Entity Resolution Policy as 'A standing procedure for deciding whether two records refer to the same entity by applying an explicit same-as criterion rather than raw token matching', so its operative form is Protocol, Workflow & Routine.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Convergent development
Present-day reach: Multi-domain
Rationale: Statistical record linkage established evidence-based decisions about whether differently rendered records refer to the same underlying entity.
Related originating lineages:
- Computer Science & Software Engineering — Database integration supplied surviving identifiers, aliases, reversible merges, and lineage.
- Library & Information Science — Authority control independently supplied governed preferred forms, variants, and cross-references.
Review resolution: Fellegi and Sunter's foundational record-linkage theory most directly establishes the same-entity decision, while Library of Congress authority records demonstrate an independent governed-identity lineage.
Attribution caveat: Statistical linkage, database integration, and authority control materially converge in the policy.
Review outcome: Researched adjudication after independent review; high confidence.
Sources consulted:
References¶
[1] Fellegi, I. P., and A. B. Sunter. "A Theory for Record Linkage". Journal of the American Statistical Association 64(328), 1183–1210 (1969). Develops a probabilistic record-linkage decision rule in which comparison-pattern probabilities are inputs to link, non-link, and possible-link classifications. registry ↩