How Veritus scores a signup: the 0-100 model and verdicts

Once the seven enrichment signals are gathered, Veritus feeds them into a LightGBM machine-learning model that produces a single score from 0 to 100 together with a verdict and a set of human-readable reason codes.

The verdict falls into one of three outcomes:

• Allow — the signup looks genuine; let it through.
• Review — the signup is borderline; send it to a human or a queue for a second look.
• Block — the signup looks fraudulent; stop it before an account is created.

You are in control of the thresholds. You decide the score at which a check is blocked and the score at which it is sent for review, so you can tune Veritus to be as strict or as lenient as your business needs.

The whole process completes in under 100ms, fast enough to sit inline with a real form submission without your visitors noticing a delay.