What coverage does ordinary split conformal prediction provide?

Question: A split conformal predictor is correctly implemented on exchangeable observations with nominal miscoverage α. Which guarantee is generally available without additional distributional assumptions? A) At every feature value x, the conditional coverage is at least 1−α B) The prediction set has marginal coverage of approximately or at least 1−α for a new exchangeable observation C) Every sufficiently large demographic subgroup has exact coverage 1−α D) The produced set is the minimum-volume set among all predictors satisfying marginal coverage Correct: B Explanation: Ordinary conformal prediction gives a marginal finite-sample guarantee over a new exchangeable test observation. Coverage may vary substantially across feature regions or subgroups. Exact universal conditional coverage is generally impossible without strong assumptions or weakened targets. Topic: advanced ML / conformal prediction / uncertainty quantification