Skip to Content
How It WorksConnections & Fingerprints

Connections & Fingerprints

Classifyre links your data in two fundamentally different ways. Confusing them leads to bad conclusions, so this page draws the line clearly.


Fingerprints: shared values, deterministic

A fingerprint connection means two assets provably share the same concrete value — the same email address, phone number, account ID, or name (with light normalisation, so [email protected] and a lowercase variant still match). There’s no guessing involved: you can always see exactly which value ties two assets together.

Every fingerprint link carries a similarity score describing how much overlap there is, and two thresholds turn that score into plain categories:

CategoryMeaning
Similar / relatedEnough shared values to be worth a look.
DuplicateSo much overlap they’re almost certainly the same underlying record or entity.

When several assets are strongly linked, they’re grouped into a cluster — treated as one entity even when they live in different systems. A cluster touching several sources is often the first sign that the same person or record has left a trail across your data estate.

Semantic similarity: meaning, not identity

Ranking’s semantic layer (see Ranking & the Semantic Layer) also compares findings by meaning — do two findings sit in similar contexts, even without sharing an exact value? This is useful for surfacing neighbours worth reviewing, but it is not proof of identity. Two documents can read alike without being about the same person, transaction, or event.

The crisp distinction: a fingerprint link says “these definitely share this exact value.” A semantic similarity says “these read alike — go check.” Fingerprints are evidence; similarity is a lead.


Tuning what counts

Fingerprinting has sensible defaults, but you can adjust it to your domain from the Fingerprints Tune panel:

ControlWhat it does
Value weightsMake some kinds of value count for more — a shared national ID is much stronger evidence than a shared city.
ThresholdsMove the bar for what counts as “related” versus “duplicate.”
ExclusionsIgnore noisy values that shouldn’t link anything, like placeholder text or a shared support address.

Changing any of these recomputes the connections across your data, so the graph and clusters immediately reflect the new rules.


Where this leads

A strong cluster or a striking connection is one of the fastest ways to open a case: its members come in as a ready-made starting point. For the full technical detail, see Fingerprints.

Last updated on