Core-depth geochemical, hydraulic and macropore data supporting Bayesian analysis of phosphogypsum-affected clay

This dataset supports the associated study of contaminant stratification in phosphogypsum-affected clay. It contains geochemical measurements from 39 core-depth observations collected from three 600-mm soil cores,This dataset supports the associated study of contaminant stratification in phosphogypsum-affected clay. It contains geochemical measurements from 39 core-depth observations collected from three 600-mm soil cores, including pH, sulfate, total phosphorus, fluoride, As, Zn, Cu, Pb and Cr. The dataset also includes 18 independently measured saturated hydraulic-conductivity observations, macropore characteristics from nine X-ray computed-tomography regions of interest, and the 39-row covariate table used for statistical modeling. Derived data include the Anion_score, Mobile_metal_score and element-specific response variables. Final Bayesian model-comparison results, mechanism-consistent pH-mitigation counterfactual outputs, and pooled leave-one-core-out TabPFN predictions and performance metrics are also provided. CT-derived variables mapped to the 39 profile rows are modeling covariates and should not be interpreted as 39 independent CT observations. The repository includes a data dictionary and offline scripts for rebuilding response scores, summarizing the locked model outputs and reproducing pooled TabPFN metrics without retraining the external modelincluding pH, sulfate, total phosphorus, fluoride, As, Zn, Cu, Pb and Cr. The dataset also includes 18 independently measured saturated hydraulic-conductivity observations, macropore characteristics from nine X-ray computed-tomography regions of interest, and the 39-row covariate table used for statistical modeling. Derived data include the Anion_score, Mobile_metal_score and element-specific response variables. Final Bayesian model-comparison results, mechanism-consistent pH-mitigation counterfactual outputs, and pooled leave-one-core-out TabPFN predictions and performance metrics are also provided. CT-derived variables mapped to the 39 profile rows are modeling covariates and should not be interpreted as 39 independent CT observations. The repository includes a data dictionary and offline scripts for rebuilding response scores, summarizing the locked model outputs and reproducing pooled TabPFN metrics without retraining the external model