Supplementary Material for "Classwise Conformal Coverage for Frozen TabPFN v2 under Imbalance"

This record contains the supplementary material for “Classwise Conformal Coverage for Frozen TabPFN v2 under Imbalance.” It provides per-seed benchmark records, calibration-budget results, dataset and split audits, reviewer-requested worst-class coverage distributions, worst-class identity summaries, leave-one-dataset-out analyses, pinned dependency specifications, source hashes, and runnable analysis code. The study evaluates frozen TabPFN v2 and comparison models on 20 public multiclass OpenML datasets under separated context, bias-estimation, calibration, and test folds. The archive does not redistribute the underlying OpenML datasets or frozen probability caches.