Sanctions & name matching

Will your screening catch the name when it's written differently?

Listed people rarely reach your payments spelled the way the list spells them. We test how your screening handles titles, initials, separators, scripts, transliteration and typos against the OFAC, UK, EU and UN lists, and show where each miss happens.

REFERENCE RUN · 22 SEP 2026 · THRESHOLD 0.70Measured
Language & cultural variants84.6% · 396
Name structure & formatting75.7% · 1,166
Compound: two difficulties at once68.4% · 171
DetectedNever retrievedRejected by ruleBelow threshold
Open-source reference matcher, self-hosted, against the UK Sanctions List.

Where a miss happens

A name can fail at three stages. Lowering the threshold reaches only one of them.

Never retrieved

The name never became a candidate, so there was no score. Underscores and native-script names often fail here.

Engine · ingestion

Rejected by rule

Scored, then thrown out by a validation rule the threshold doesn't control. Chinese and Korean two-token names fail here.

Engine · matching logic

Below threshold

Scored too low to alert. The only stage your configuration reaches, and every recovered match costs alerts.

Configuration

Where a miss comes from the list itself, such as a published alias or native-script form your data doesn't hold, the report marks it Data and keeps it apart from the matching results.

Six test categories

c. 50,000 name variants and c. 25,000 ordinary names in every baseline.

01

Name & data handling

Did the system receive and parse the name correctly?

Truncation, missing or wrong fields, delimiters

02

Name structure & formatting

Is this still the same name, just represented differently?

Name order, initials, titles, suffixes, separators such as dots and hyphens

03

Linguistic variants

Does it understand how the same name changes across languages?

Transliteration, original script, morphology, name equivalents and nicknames, diacritics, spacing and joining

04

Character errors & obfuscation

Can it survive imperfect or deliberately altered characters?

Typos, keyboard errors, OCR errors, look-alike characters

05

Compound & stress

Does it still work when several problems happen at once?

Several of the above in one name

06

False positives & efficiency

Does it know when not to match?

Ordinary names by geography, and near-namesakes who share a listed name

Sample report · linguistic variants · October 2026

Every name the list publishes was found. One in three forms it doesn't publish was missed.

469 real variants of listed names, the way they actually arrive: transliterated, in their own script, as nicknames, in another country's spelling. Run on Escalate's reference engine, built on open-source matching, at a 0.70 threshold.

Effectiveness76.8%360 of 469 name variants found, at threshold 0.70
Efficiency96.0%240 of 250 ordinary names left alone
Exact-name checks110 / 110Controls passed, so the test is sound

Catches more, alerts less

BETTER ENGINE88%92%96%100%60%70%80%90%0.600.70 · this run0.90Ordinary names left alone →Variants found →

Each dot is a threshold. Lowering it from 0.70 to 0.60 finds 20 more variants but nearly triples false alerts on ordinary names, from 10 to 27. Reaching the shaded corner takes a better engine, not a different setting.

Where the 469 variants ended up

Found · 360Not found · 21Rule · 2Scored too low · 86

Every form the list publishes was found: 167 of 167. One in three forms it doesn't publish was missed. A name with its own script attached was found 2 times out of 40; everyday nicknames such as Dima for Dmitry, 1 of 27.

By naming tradition

South Asian87.2%Own script next to the Latin name: 0 of 5
Arabic80.0%DIN 31635 scholarly standard: 0 of 2
Chinese73.8%Pinyin with tone numbers: 0 of 3
Korean68.2%Hangul, South Korean spelling: 0 of 2
Persian57.1%Turkish form of the given name: 0 of 2
Russian & Ukrainian55.7%German-style romanisation: 0 of 9

Real submissions, real scores

Ordinary forms of a listed name, at a 0.70 threshold.

SubmittedScoreOutcome
I Leontievna DIANOVA0.697Below threshold
Anton.Vyacheslavovich.KRASOVSKY1.000Detected
Anton_Vyacheslavovich_KRASOVSKY–Never retrieved
Anton Vyacheslavovich KRASOVSKY (КРАСОВСКИЙ Антон Вячеславович)0.579Below threshold
Wei ZHAO0.900Rejected by rule
PHOENIX COMPANY–No alert

Listed on the UK Sanctions List as Irina Leontievna DIANOVA, Anton Vyacheslavovich KRASOVSKY, Wei ZHAO and PHOENIX LLC. Adding a name's own script in brackets, as many payment messages do, pulled a certain match below threshold.

Also in every programme

Recall is half the picture.

False positives by corridor

Every should-match class is paired with a should-not-match collision from the same naming tradition, so you see what each extra catch costs in alerts, corridor by corridor.See the false positives sample →

Drift between tests

Lists change, vendors push updates and thresholds get tuned. A fixed core of cases shows when performance moved, and what moved it.

The lists you screen

OFAC, UK, EU and UN, weighted to the corridors you serve. For data providers we also test the list data itself: 33,000+ records and about 105,000 published names, original script, identifiers, vessels and aircraft, cross-list links and delistings, plus how fast new designations appear as an add-on.

Run this on your engine

A blind set of about 500 names, weighted to your corridors.

The figures on this page describe an open-source reference matcher, not your system. Your report comes from the same test classes, run through your own screening.

  1. 20 minutes to agree corridors, PEP scope, and what you return: hit or no hit, and the score.
  2. You receive test IDs and names only, with no expected results.
  3. You run the file and send back the results.
  4. You get the report on your own system, with each miss split by engine, configuration and data.