Definition
Conformal prediction is a statistical framework that turns any model’s raw outputs into calibrated prediction sets — an interval or a set of labels that is guaranteed to contain the true answer at a user-chosen rate (say 90% of the time), regardless of the underlying model or data distribution. It earns that distribution-free guarantee by calibrating a nonconformity score on held-out data, which is what lets a system bound its error or violation rate without assuming the model is correctly specified.