Mahal Systems OpenSanctions Pairs
Fits compliance screening or know-your-customer agents that match entities against sanctions lists. Borrow its pairwise labeled format from 293 sources across 45 jurisdictions and its rule-based baseline (91.3% F1) as a floor.
What it asks the agent to do
- decide whether two sanctions or watchlist records refer to the same person or company
- match names across languages and writing systems
- compare records with partial dates, aliases and identifiers
What to borrow from it
baselinecase template
Caveats
Pairwise matching is already near ceiling (GPT-4o 99.0% F1) so it will not separate models; records are real named persons and organizations; no dataset or code link was found on the abstract page and the data license is not stated there.
Details
| Workloads | data analysis, finance, government, security |
|---|---|
| Interaction | single turn |
| Environment | static |
| End state checked | No |
| Tools | none |
| Grading | programmatic check |
| Size | 755,540 |
| Data | real public |
| Personal data | yes |
| License | unknown |
| Harness | none |
| Status | saturated |