How we screen names

In short

• We fetch 10 official sanctions lists from their publishers every day.
• The matching is deterministic and uses no AI or language model: the same names and the same list data give the same result.
• Every match is a possible match for you to review. The service does not decide for you.
• Every file report records the data date of each list, the method version and the SHA-256 fingerprint of your file.

The 10 lists and how we keep them current

We fetch each list from its official publisher every day. The table links the official source of each list.

If a download fails, we keep the last good copy of that list, and its data date stays the same. The report marks a list as out of date when its data is more than 48 hours old.

Every report shows the data date of each list. The current dates are also at /api/lists/status (JSON).

The DFAT Consolidated List check page explains the Australian list and how to check a client list against it.

How we prepare names

• Upper and lower case, accents and punctuation are made uniform, so “José” and “JOSE” are the same word.
• In company names, legal forms such as Ltd, GmbH and S.A. do not count as name words.
• Names in other scripts are compared with the Latin spellings that the lists use: Cyrillic (BGN/PCGN), Greek (ELOT 743), Armenian and Georgian (BGN/PCGN), Korean Hangul (as the lists write North Korean names), Chinese characters (pinyin), and Arabic and Persian script.
• We compare your name with the main name and with the other names (aliases) of each list entry.

How the name score works

The “About this screening” section of every report explains the score in the same way.

Name score (0-100%): SoftTFIDF word alignment (Cohen, Ravikumar, Fienberg 2003). Each word of your name pairs with at most one word of the list name. A word that is rare across the lists weighs more than a common one (IDF).

• Jaro-Winkler: word similarity.
• Double Metaphone: words that sound alike, when their spelling is already close.
• Damerau-Levenshtein: edit distance for spelling variants and typos; a swap of two letters counts as one edit.

The same input and the same list data give the same result.

You can try the method with up to 10 names in any 24 hours for free, in the free check on the home page.

Score bands

• Strong name match: 90% or more.
• Possible name match: 60% to 89%.
• Below 60%: not reported.

Country and date of birth are separate checks per field: agrees, differs, not in your file, or not in the list entry. They never remove a match. They change the order in which the report shows the matches: a match with a value that differs, and no value that agrees, is shown lower.

One rule changes the band. When every date of birth in your file is more than 2 years from every date of birth of the list entry, and the report can read all these dates, a strong name match is shown as a possible match (89% at most). The report then also gives the name score before the rule.

What a match means

A match is not a finding. It means that a name in your file is similar to a name on a list. Compare the other details, such as the date of birth, the country and the identifiers of the list entry. Then record your decision on the decision line of the report.

What we do not check

• Who owns or controls a party, for example under the OFAC 50 percent rule.
• Lists of politically exposed persons (PEP).
• Adverse media (news reports).
• Lists other than the 10 lists above.
• No AI or language model takes part in the matching.

Known limits

• A name can have more than one correct Latin spelling. A true match can be missed when your spelling and the list spelling differ too much.
• Names in some scripts, such as Hebrew, Thai or Japanese kana, are compared as written.
• A name is reported only when its name score reaches 60%.
• A list may not hold every person or entity that is subject to a sanctions measure.

These limits are also in section 4 of the Terms of Service.

What each report records

The report of a file check records:

• the SHA-256 fingerprint of your file
• the data date of each list
• the snapshot version of the list data
• the method version (the PDF shows it as “SanctionListCheck matcher 2.11”)
• the score thresholds (90% and 60%)

An API screening record holds the snapshot version, the data date of each list and the method version.

Keep the report and your file. If OFAC rules apply to a transaction, they ask you to keep its records for at least 10 years after the transaction (31 CFR 501.601). If US export rules apply, BIS asks for 5 years from the export or the end of the transaction, whichever is later (15 CFR 762.6). We delete your file and your report 24 hours after the upload, so download the report to keep it.

Method changes

Each change to the matching gets a new method version. The report names the version that screened your names. The table lists the changes since version 2.4 (dates in UTC).

More about the service is on the About page.