A method for identifying and linking records across two or more datasets that refer to the same real‑world entity by comparing persistent identifiers, combinations of attributes, or similarity scores; implemented deterministically when authoritative unique identifiers are available and probabilistically when matches are inferred from noisy or incomplete attributes.

Principle

Linking decisions follow from the evidence supporting identity: exact identifier equivalence yields high-confidence matches, while attribute similarity and contextual rules produce probabilistic matches that trade false positives against false negatives.

Demonstration

Illustrative scenario → Two institutional catalogs list 'J. Smith' with matching birthdates and one record contains an institutional identifier. Recognition → The identifier match links records deterministically; the name+birthdate pair is scored and flagged for review. Action → Deterministic links are merged automatically; probabilistic links are reconciled after clerical review. Consequence → A consolidated view of the individual's holdings is produced while uncertain links are documented and auditable.

Misapplication

Treating a single shared attribute (e.g., a common name) as proof of identity. This confuses attribute equality with entity identity and increases false‑positive links.

Consequence

Correct reconciliation enables integrated services, consolidated analytics, and reliable provenance across systems; incorrect links propagate erroneous aggregations, misattribute resources, and reduce trust in derived services.

Reversal

When a high-quality, persistent unique identifier (for example an authoritative national identifier or verified researcher ID) exists and is accessible, reconciliation reduces to deterministic matching; conversely, legislation, hashing, or strong pseudonymization that removes identifiers can prevent reconciliation or require privacy‑preserving linkage techniques.

Boundary

Clearly within: linking person records across institutional repositories using ORCID or matching multivariate identifiers. Boundary case: matching records with similar names and dates where cultural naming conventions or incomplete dates create ambiguity. Clearly outside: pure entity recognition inside text (NER) without record linking; or superficial similarity grouping that does not produce cross‑dataset links.

Semantic Tension

Privacy ↔ Utility — linking increases operational value and analytical power while raising risks to individual privacy and requiring governance controls.

Synthesis

Entity reconciliation is an evidence‑driven process whose reliability depends on identifier quality, attribute richness, and explicit thresholds; practical systems therefore combine deterministic matches, probabilistic scoring, human review, and provenance recording to manage trade‑offs between scale and accuracy.