Correlation let you down, so you escalate. You fit OLS on all 493 features and the training \(R^2\) comes back at \(1.0\). For about thirty seconds you are a hero. Then you run the model on a country it hasn’t seen and the test \(R^2\) is \(-208\) — worse than predicting the mean. This chapter walks you up to the cliff at \(p = n\) and then over it.