Skip to content

Understanding results

Field Meaning
hit true if at least one match scores at or above threshold.
threshold Your project threshold (0–100) at the time of screening.
matches Matches at or above the threshold, highest score first.
sources Every source searched and the list version used.
model_version Version of the matching pipeline, for audit.
alert_id Alert created for review when hit is true.

Each match has a final score from 0 to 100 and a score_breakdown:

  • retrieval (0–1): similarity found by the fuzzy phonetic retrieval stage.
  • verifier (0–1): the AI verifier’s estimate that both strings name the same person.
  • date_of_birth: outcome of the date comparison (exact, partial, near, mismatch, not_compared).
  • single_token_penalty: true if either name is a single word. Mononyms can match many people, so they are penalised unless the date of birth confirms the match.

matched_name and matched_name_type (primary, alias, weak_alias) show which name of the list entry produced the match.

entity contains the list entry as published: names, aliases, dates of birth, nationalities, sanctions programmes, listing date and remarks. list_version and record_version identify the exact version of the entry, so a result can be reproduced later, even after the list changes.

The threshold balances recall against false positives. Lower values return more potential matches. You set one threshold per project. Pick it using your own test set of known positives and negatives, and revisit it when model_version changes.

A hit is a potential match, not a confirmation. Typical handling:

  1. Pause the onboarding or payment.
  2. Route the case to an analyst, who compares secondary identifiers.
  3. Record the decision. Confirmed false positives can be whitelisted for that external_id, so they do not alert again until the list entry changes.