How to test sanctions screening: coverage, matching, and workflow

On this page

Test the complete control, not one search

A screening integration includes inputs, source selection, matching, outcomes, review, monitoring, and evidence. Passing a single known-name search does not prove that the complete control works.

Create a test plan around the organization’s actual use cases and approved scope. Keep synthetic fixtures separate from production customer records, and label invented identities clearly so they are never presented as real designations.

Build a labeled fixture set

Include exact and variant names, common names, aliases, partial dates, organizations with similar trading names, mismatched identifier issuers, and supported non-person entity types. State the expected candidate and the evidence supporting that expectation.

Separate positive retrieval fixtures from identity-decision fixtures. Returning a candidate is not the same as confirming the identity. Use the name-matching guide when designing explainable variations.

Exercise coverage and contract failures

Test invalid subject data, a missing required source, unsupported source/entity combinations, invalid credentials, rate limits, timeouts, malformed responses, and expired evidence. Every failure should reach a visible incomplete or error path, never a fabricated no-match result.

For retries, submit the same payload with the same idempotency key and then verify that a changed payload is not incorrectly replayed as the original operation. Use current error and idempotency contracts.

Measure retrieval quality without hiding tradeoffs

Measure missed expected candidates and unrelated candidates separately on a representative labeled set. Record the source and matching-policy versions used. Changing a threshold may reduce noise while increasing misses.

A test set cannot prove perfect real-world recall. Document what populations, scripts, formats, and cases are absent. Avoid presenting one benchmark percentage as a universal guarantee.

Test people and evidence as well as code

Have an authorized analyst review a candidate, record supporting and conflicting facts, request additional evidence, and complete required independent review. Then ask another reviewer to reconstruct the case from the retained record.

Test a changed source record, reopened candidate, stopped monitor, and failed rescreen. Acknowledging a notification should not accidentally close the linked investigation.

Use release gates that match the control

Define which failures block launch and who can authorize exceptions. Retain the test results, fixture versions, source versions, known limitations, and remediation owners. Repeat relevant tests after source, policy, parser, or workflow changes.

Use SanctionsKit’s sandbox for integration mechanics. A synthetic sandbox pass does not validate the freshness or identity accuracy of live OFAC data; production validation requires an authorized, carefully scoped process.