A risk rating, often called a risk score, is a number used to guide a decision under uncertainty. Its meaning depends on the decision: a card transaction may receive a fraud-risk score, while a person may receive a score used in criminal-justice decisions. The number does not explain itself; its inputs and consequences matter. section 4.1.2 section 10.2
A card purchase before Christmas
Before authorizing a purchase, a fraud-prevention system uses cardholder and merchant profiles and other data to generate a score. It approves transactions only when the score is below a threshold. An unusual purchase might be suspicious in July but ordinary before Christmas, so adjusting scores or thresholds to the circumstances can reduce false positives. section 4.1.2
Profiles can draw on behavior over time, locations, interactions among entities, and comparisons with peers. Sharing profile information among agents can help reveal group patterns and emerging fraud. No scoring formula or procedure for checking a score’s accuracy is specified here. section 4.1.2
The stakes change when scores affect people’s liberty. Criminal-justice risk scores described on a 0–10 scale can draw on education, employment, neighborhood, family or friends’ records, and answers to moral questions. They can influence bail, sentencing, incarceration, and parole. Such uses call for scrutiny of whether the inputs and resulting decisions are rational, objective, and free of bias—not acceptance of a number as a verdict. section 10.2
