Homoscedasticity: when the noise grows with the data

Heteroscedasticity never touches your coefficients. It quietly breaks your standard errors — and every t-test and confidence interval built on top of them.