The Skew

A medium ai_coding interview practice problem on DataDriven. Write and execute real ai_coding code with instant grading.

Domain

ai_coding

Difficulty

medium

Seniority

L5

Problem

ML team's fraud model trains at 0.84 AUC offline and tanks online. We isolated it to feature `user_orders_last_24h` , offline backfills compute it correctly, but the online value at scoring time is consistently higher. Run tests/test_feature_parity.py to see the divergence on a known prediction. Find the time-leak in the offline pipeline (features/orders_24h.py) and fix it. Defend your fix: what's the upper bound on staleness you've now committed the team to?

Summary

Lab-perfect. Production-broken.

Practice This Problem

Solve this ai_coding problem with real code execution. DataDriven runs your solution and grades it automatically.