What does standard conformal prediction guarantee?

Question: Under exchangeability and correct implementation, what does ordinary split conformal prediction primarily guarantee at miscoverage level α? A) Conditional coverage of at least 1−α for every possible feature vector x B) Marginal coverage of approximately or at least 1−α over a new exchangeable observation C) Coverage of 1−α separately within every demographic subgroup D) The shortest possible prediction set among all valid uncertainty methods Correct: B Explanation: Standard conformal validity is marginal over the randomness of a new exchangeable example. It does not generally guarantee exact conditional coverage for every x or every subgroup, and it does not guarantee optimal set size. Topic: advanced ML / conformal prediction / uncertainty