Record Linkage

Record linkage establishes identity correspondence between records from separate datasets. Its useful output is a crosswalk that keeps both source identifiers and the evidence for each accepted relationship.

In a fictional venue example, roster key A17 and event-system key B04 may describe the same hall. Retain roster:A17 and events:B04 instead of replacing both with an unexplained name. If B09 shares the booking email but refers to an annex, that extra candidate remains visible until the owner resolves it.

An ordinary exact join uses an already trusted relationship between keys. Linkage is the earlier decision about whether those keys correspond at all. Applying a join to unreviewed candidates can multiply rows or attach activity to the wrong entity even when the SQL runs successfully.

The cross-source linkage workflow starts by defining record grain and permitted cardinality. A one-to-many relationship can be legitimate when one source contains historical identifiers. It can also reveal an ambiguous match. The policy and evidence determine which interpretation is allowed.

Entity resolution is the broader identity problem. Dedupe is one library to evaluate for candidate generation. Whatever tool you choose, separate accepted links, unresolved candidates, and unmatched records. Record the human approval, source versions, and subsequent join checks so corrections can be traced to the crosswalk version that produced them.