Explainable Name Screening Against the OFAC SDN List: An Open Implementation and Reproducible Evaluation

Organizations in the United States must not deal with parties on the Specially Designated Nationals (SDN) list maintained by the Treasury's Office of Foreign Assets Control (OFAC), so lenders, payment firms and many other businesses screen every customer name against it. Screening is a fuzzy-matching problem, yet most screening logic in use is either proprietary or undocumented, which makes alerts hard to explain to analysts and examiners. We present OfacMatch, an open-source screener that follows the two-technique method OFAC describes for its own search tool, combining whole-name Jaro-Winkler similarity with name-part matching that uses Soundex, and that returns a part-by-part explanation for every potential match. We add a guard that prevents a shared first name from carrying the whole-name score, and we flag date-of-birth and country conflicts instead of hiding matches. On the real SDN list (19,484 entries), with 4,255 synthetic but realistic variations of 600 listed parties and 2,000 ordinary names as controls, OfacMatch finds 99.8% of variations at a threshold of 85. The guard lowers the alert rate on ordinary names from 63.2% to 47.9% at that threshold, at a cost of 0.2 percentage points of recall. The implementation, the evaluation harness and its fixed random seed are public, so every number in this paper can be reproduced.