conformal/

Wrappers that convert a fitted point predictor into a set or interval predictor with a distribution-free, finite-sample coverage guarantee.

• Split conformal: one model fit, needs a held-out calibration split.
• CV+ / jackknife+: cross-fitting instead of a held-out split; uses all the

data at the cost of k fits.

• APS / RAPS: adaptive sets that grow on ambiguous inputs.
• Mondrian: a separate threshold per class or group, for conditional rather

than merely marginal coverage.

• CQR: conformalised quantile regression, for heteroscedastic intervals.
• Venn-Abers: probability intervals that report their own calibration

uncertainty.