Conformal-Coverage-Failure-Risk-Dataset

Simulated conformal-prediction coverage-gap trajectories for 180 configurations across 6 methods (vanilla split conformal, weighted conformal, jackknife+, RAPS, label-shift-corrected conformal, adaptive conformal inference) and covariate/label distribution shift, grounded in the 2019-2023 distribution-free uncertainty quantification literature. Forward-looking task predicts long-run coverage failure from a quick post-shift check.

About Dataset

Simulates whether a conformal method's coverage guarantee survives distribution shift — including a genuine method-shift-type mismatch (a covariate-shift-corrected method fails under label shift and vice versa) and adaptive conformal inference's documented transient overshoot right after shift onset, recovering over subsequent rounds. 180 configs across 6 methods, 3 shift types. Ten sources (2005-2023) including the foundational conformal framework, conformal prediction under covariate shift, adaptive conformal inference, and conformal prediction beyond exchangeability.

Purpose of Dataset

Lets A Team Monitoring A Deployed Conformal Predictor Estimate, From A Quick Check Right After Noticing A Shift, Whether Their Method Will Recover Or Stay Broken.