Engineering

How we build fraud models that hold up in production

Notes from our engineering team on evaluating, calibrating and serving fraud models trained on each merchant’s own data.

When fraud history knows the future: the two clocks inside a fraud feature

How future dispute data leaks into fraud features, why lookback windows don’t fix it, and how to rebuild history around outcome event times.

Bo HuangNov 12, 2026 · 8 min read

EvaluationNov 11, 2026
Measuring False Declines You Can’t See: What a Fraud-Vendor Switch Reveals
How to measure the false declines your fraud screen hides, using vendor-switch data from 30,612 high-risk payments as a natural experiment.
Qing Wang18 min read