Trust score

A single number summarizing how likely an identity is genuine.

A service might approve applicants with high scores automatically, send middle scores to a human reviewer, and decline low scores, so the number turns many separate checks into one decision.

A model output aggregating document, biometric, device, and network signals into a calibrated risk score. Calibrated means the number reflects real-world likelihood, so a given score carries roughly the same level of risk from one month to the next.

A trust score is typically built like this:

1. The system collects signals from the session, such as document checks, a face match, a liveness result, the device, and the network.
2. A model weighs the signals and produces a score.
3. The service compares the score with thresholds it has set.
4. The case is approved, declined, or routed for review based on where the score falls.

The inputs come from several places:

• Document signals, such as whether the ID looks genuine and its data is consistent.
• Biometric signals, such as the face match score and the liveness result.
• Device and network signals, such as whether the device is an emulator or the IP address comes from a datacenter.
• Behavioral signals, such as how the person moves through the flow.

A trust score sits next to terms that are easy to mix up:

• Risk signal is a single data point that feeds the score.
• Decisioning engine is the component that applies thresholds and policy to produce approve, decline, or review.
• Rules engine applies explicit if-then logic, which can run alongside or on top of the score.