01
Same or different
Make the identity decision explicit instead of relying on one field.
07 / Data
Identity Resolution
Match records with an explicit review state instead of treating a similar name as proof of identity.
The pair
Use uncertainty to prevent false merges: 0.78record A
record B
mia@northstar.io
mia@northstar.io
Northstar Labs
Northstar
+1 415 555 0192
—
Finance lead
Operator
Quality controlpossible match · review
The pair
The name is only one signal
Entity resolution works best when the decision can explain agreement, conflict and missing evidence.
02
Field agreement
See which fields support the match and which ones conflict.
03
Review state
Keep ambiguous pairs separate until a person or policy resolves them.
The pair
01
reviewEntity resolution works best when the decision can explain agreement, conflict and missing evidence.Find candidates
Use deterministic matching or embeddings to create a manageable candidate set.
02
reviewEntity resolution works best when the decision can explain agreement, conflict and missing evidence.Compare fields
Pass names, addresses, domains, IDs and other identity evidence together.
03
reviewEntity resolution works best when the decision can explain agreement, conflict and missing evidence.Respect policy
Describe which evidence is strong, weak or disqualifying for your domain.
04
reviewEntity resolution works best when the decision can explain agreement, conflict and missing evidence.Avoid blind merges
Route low-confidence matches to a review queue instead of forcing a result.
Compare two records
Measure false merges before increasing automation
A missed match is often recoverable. A wrong merge can corrupt every downstream record.
Compare two records
POST /v1/resolve{
"state": {
"record_a": { "name": "Acme Inc.", "domain": "acme.io", "city": "Oakland" },
"record_b": { "name": "ACME Incorporated", "domain": "acme.com", "city": "Oakland" }
},
"questions": {
"identity": { "type": "choice", "instructions": "How should these records be treated?", "criteria": { "same": "same entity", "different": "different entities", "review": "not enough evidence" } },
"domain_conflict": { "type": "noul", "instructions": "Is the domain conflict material?" }
}
}FAQ
Questions about identity resolution
Does this replace deterministic matching?+
No. Use deterministic keys and candidate generation first. Jev can help review the ambiguous pairs that remain.
Should the output be binary?+
Usually not. Same, different and review gives your pipeline a safer middle state.
How do I prevent false merges?+
Define strong evidence, keep conflicts visible and require a higher threshold for irreversible merges.
