A customer churn dataset contains a cancellation timestamp that is populated only after the customer leaves. Offline validation is excellent, but predictions at signup fail. What is the most likely issue?

1. The training instance is too small, regardless of feature availability.
2. The model uses a feature unavailable at prediction time, creating target leakage.
3. A larger validation set containing the same future feature solves the issue.
4. The model needs the same cancellation timestamp supplied for future customers.