Build a real-time ATO model

End-to-end ML case for ATO: Design a real-time model to detect Venmo account takeover at payment authorization time. Requirements: P99 scoring latency < 20 ms; features available online via a feature store/Redis; labels from confirmed ATO/chargebacks with a median delay of 45 days; daily traffic 1M+ tx. Tasks: A) Precisely define the positive label (ATO) and the negative set; discuss positive–unlabeled learning and how to construct reliable training data with delayed/noisy labels. B) Propose features across device, IP, behavior, network, and account age; for at least 3 features, specify leakage risks and how you’d time-travel-proof them. C) Select a model family (e.g., gradient boosting with monotonic constraints) and justify; include calibration (Platt vs. isotonic) and how you’ll maintain calibration across account-age cohorts. D) Describe an offline evaluation protocol (time-based split, label latency handling, group-aware CV) and online validation (shadow mode, interleaving with rules). E) Outline drift/adversary monitoring and automated retraining triggers (e.g., PSI thresholds, population/conditional shift tests); F) Explain how to combine ML score with deterministic rules via a policy engine to meet business constraints (e.g., block, step-up auth, allow) and how to set per-segment thresholds to hit target FP/FN budgets.