Precision versus recall, on one dial.
Screening teams cannot have it both ways. A looser filter catches more true matches but floods analysts with false positives; a stricter one is lighter to work but lets real matches through. This model makes that trade-off something you can feel — always from the defender’s side.
Tune the match threshold
SYNTHETIC — TRAINING ONLYEvery name on a payment is compared to the watchlist, and the engine returns a similarity score. Anything at or above the threshold becomes an alert an analyst must clear. Drag the threshold and watch what it does to the two numbers that fight each other.
At 60% every real match is caught — but 5 false positives must be cleared by hand to get there. That is the cost of high recall.
See every candidate and how it is classified now
| On the payment | Watchlist entry | Similarity | Truth | Now |
|---|---|---|---|---|
| Arjun Mehta | Arjun Mehta | 98% | Listed party | True alert |
| Riya Sharma | Riya Sharmaa | 90% | Listed party | True alert |
| M. Kabir | Mohammed Kabir | 72% | Listed party | True alert |
| Asha Traders Ltd | Asha Trading Limited | 68% | Listed party | True alert |
| Nadia Petrov | Nadya Petrova | 63% | Listed party | True alert |
| John Smith | John Smith | 95% | Not listed | False positive |
| Maria Garcia | Maria Garcia | 94% | Not listed | False positive |
| Meridian Bank | Meridian Capital | 70% | Not listed | False positive |
| David Cohen | Daniel Cohen | 66% | Not listed | False positive |
| Li Wei | Li Wei | 60% | Not listed | False positive |
| Cassia Bank | Acacia Bank | 55% | Not listed | Correctly cleared |
| Kabir Ahmed | Kabir Ahmad | 50% | Not listed | Correctly cleared |
| Priya Nair | Pooja Nair | 42% | Not listed | Correctly cleared |
| Nordbank AB | Sudbank AB | 38% | Not listed | Correctly cleared |
This is the defender’s tuning problem — how to set controls so real matches are caught without burying analysts. It never shows how to avoid detection.