Practical guide
Sanctions screening false negatives: where missed matches arise
Understand missed sanctions candidates across identity inputs, coverage, source freshness, matching, and workflow failures, with practical tests for each layer.
On this page
A missed candidate can have more than one cause
A false negative in a defined screening test means an expected relevant candidate was not returned. In a real process, an apparent miss can arise from the wrong subject, missing coverage, stale data, retrieval behavior, or a result that never reached review. Diagnose the layer before changing a threshold.
A direct-list search also has an intentional scope limit. An unlisted company affected by ownership restrictions is not proof that a name-matching algorithm failed to retrieve a published direct listing that never existed.
Check identity and coverage first
Confirm that the submitted subject was the correct legal person or organization, with reliable known facts and the correct entity type. Then verify that the required source was selected, available, and included in the completed result.
A trading name in place of the contracting legal entity or a missing source selector can explain a gap that a more permissive fuzzy matcher would not fix.
Check the source version and timing
Record the authority publication context, ingestion time, active screening version, and result time. A successful download does not necessarily mean a validated new snapshot became active. A scheduled rescreen can run against an unchanged snapshot.
Use the data-freshness guide to separate these clocks. Do not promise that any provider sees every authority change instantly.
Test the matching behavior deliberately
Use labeled cases for aliases, name order, spelling changes, partial dates, identifier handling, and supported scripts. Preserve the exact input and source record. Compare behavior before and after a change without quietly altering the expected answer.
A score is not a calibrated probability. Evaluate matching thresholds through measured retrieval tradeoffs rather than a universal “safe” number.
Look for workflow failures after retrieval
A candidate may have been generated but hidden by a stale dismissal, dropped webhook, failed queue handoff, or incorrect case-level approval. Test these paths with synthetic cases. An acknowledged monitoring notification is not necessarily a resolved investigation.
Keep technical failures visible and never default missing response data to an empty candidate list. This is a software safety property that can be tested independently of source accuracy.
Document the limit of the conclusion
A representative test set helps find defects and compare changes. It does not prove that no real-world identity will ever be missed. Record coverage gaps, unsupported populations, unresolved issues, and separate ownership or activity controls.
Use screening validation and ownership guidance together. A useful effectiveness claim states the population, source versions, test method, and remaining limitations.