Introduction
"Our model is 87% accurate" is incomplete — on a different test set it might be 85% or 89%. A confidence interval reports a range of plausible values together with how confident we are in the procedure. Reporting intervals instead of single numbers is how you tell real improvements from noise.