Identifiers classify
Heuristicidentifiers_classifyBest-effort — fuzzy, approximate, or inferred; check the confidence score.
Classify a string as a known ID format (SSN, EIN, TIN, ITIN, NPI, UUID, ISBN, etc.) and return ranked candidates with confidence scores. Format-based only — does not run checksum validation. For checksum validation use the dedicated validators (healthcare_npi_validate, healthcare_dea_validate). Processing multiple inputs? Use `identifiers_classify_batch` to do up to 1000 in one call.
Parameters
valuestringrequiredString to classify by format.
Call it
curl -X POST https://api.snipget.ai/v1/identifiers/classify \
-H "Authorization: Bearer sk_live_..." \
-H "Content-Type: application/json" \
-d '{"value":"123-45-6789"}'The Batch tab sends up to 1000 inputs in one call via /v1/identifiers/classify/batch.
Example response
{
"input": "123-45-6789",
"candidates": [
{
"type": "SSN",
"confidence": 0.95,
"reason": "matches XXX-XX-XXXX format"
}
],
"top_match": {
"type": "SSN",
"confidence": 0.95,
"reason": "matches XXX-XX-XXXX format"
}
}FAQ
What does this tell me?
It guesses what kind of identifier a string is (SSN, EIN, NPI, ITIN, UUID, ISBN, and a few more) by its shape, and returns ranked candidates with a confidence score for each.
Does it confirm the number is real or valid?
No. It only matches the format, not the math. A nine-digit string can look like an SSN or an EIN without being a genuine one. For a real checksum check use the dedicated validators (Luhn, NPI, ISBN, and so on).
Why can it return more than one answer?
Some formats overlap. Plain nine digits could be an SSN, EIN, or TIN, so it lists all of them ordered by confidence, with the best guess as top_match.
Is the answer stable?
Yes. It is deterministic: the same input always returns the same ranked candidates. The classification itself is a best-effort shape heuristic, and the confidence numbers reflect format ambiguity, not randomness.
What if nothing matches?
You get an empty candidate list and top_match of null with confidence 0, not an error.