Possible Duplicate

What It Is

A possible duplicate is a question about two contacts or two organizations: do these records represent the same identity? It gives staff a concrete suggestion to investigate while leaving both records available for ordinary work. A second registration, a staff-created attendee or an integration can create the resemblance that detection finds. The pair is the unit of evidence; a group gathers connected pairs for review.

Each suggestion carries a score, a confidence band and reasons that explain the matching and conflicting information. The result also identifies how the pair arose. Computed suggestions can disappear when the resemblance no longer qualifies. Manual confirmations and declined portal suggestions retain a human decision that ordinary rescoring does not erase. A stored score is therefore evidence about resemblance, rather than a substitute for knowing the people involved.

What It Is Not

A suggestion does not combine records, remove a person or change either record's contact status. It grants no additional merge authority. A dismissal concerns the named pair only; it does not exempt either record from comparison with everyone else. Configuration determines the response to a match, while the evidence model determines its score.

How It Is Used

Reviewers inspect groups from Data Quality or investigate suggestions on a record page. They read the reasons, check the identity and dismiss false matches. An administrator can launch a merge plan when the records really belong together. Manual confirmation is supported by the service, although its planned console entry points remain incomplete. A shared queue address helps colleagues return to the same review location.

New strong evidence can make an earlier dismissal worth reconsidering. Reviewers then receive an explanation of the reopening rather than a silent reversal. Good review distinguishes shared household or employer details from personal identity evidence. An organization's linked address or phone does not make its employees the same person.

Its Facets

Match keys are normalized details used to locate plausible comparisons. Common values can be excluded to avoid filling the queue with weak associations. Scores and bands express the fixed evidence model, including ceilings for contradictions. Reasons translate that evidence into readable sentences and mask private contact values.

Origin distinguishes computation from manual or portal decisions. Status and dismissal retain open, dismissed or confirmed standing together with the reviewer and reviewed evidence. Groups collect connected records by status, preserve useful identity through changes and report when their size is bounded.

Duplicate settings belong to this concept rather than being another product object to understand. Their record type and match pattern select the applicable evidence. Their action determines review, warning or blocking. Minimum confidence supplies a floor, position supplies precedence and active status enables evaluation. The portal switch is stored without controlling portal matching today. Draft checks expose up to five matches before creation. Override records retain the author and reason for authorized exceptions. Re-checks refresh existing evidence after a setting changes; configuration alone does not immediately reevaluate the queue.

How It Relates

Entities owns the contacts and organizations named by a suggestion. Data Quality owns their comparison and review decisions. A duplicate group commonly starts a merge plan, but planning can also begin with manually selected records. Successful merging removes absorbed-record suggestions and carries relevant outside dismissals to the survivor.

What It Could Become

The product should substantiate confidence bands with calibration against synthetic examples and current review decisions. Console confirmation controls and re-check management remain unfinished. Responsibility-based attention is declared but held. Completing the import-completion connection would let imports receive a deliberate final detection pass instead of current per-record processing.