Conformal uncertainty for deformable registration under acquisition shift: code, pre-registration, patient-level splits and result tables

Analysis code, patient-level split definitions with the patient-leakage audit, the pre-registration with all dated amendments, the DF-WCQR method specification, engine manifests, per-pair conformal scores, evaluation tables, residual-quantile predictor weights and the invalidated runs with their diagnostics, for the manuscript "Conformal uncertainty for deformable image registration under acquisition shift: coverage transferability, geometric relevance and downstream utility in abdominal CT" (submitted, 2026). The study calibrates an appearance-residual uncertainty for a pair-wise abdominal CT registration engine with conformalized quantile regression and measures three properties separately: coverage under contrast-phase and vendor shift (including a density-ratio weighted variant, DF-WCQR, for held-out vendors), relation to landmark target registration error on external intra-patient data, and the effect of uncertainty-guided exclusion on label propagation. Data: AMOS22, Learn2Reg AbdomenCTCT and an abdominal vessel-bifurcation landmark dataset, all public. Not included: images, labels, pseudo-label volumes or displacement fields; obtain the public datasets from their original sources. No institutional data were used in any form. Birth dates and acquisition dates used to infer patient groups have been replaced by group numbers. Code is MIT-licensed (LICENSE file inside the archive); documents, tables and model weights are CC BY 4.0. See README.md in the archive for the layout and the mapping to the manuscript's tables and figures.