Prioritizing soil Pb verification targets under spatial outer testing for regional monitoring

Regional soil Pb monitoring must decide which samples to verify first under a fixed budget. We distinguish contamination status, screening status, and verification priority; the model ranks verification targets using labels defined by prespecified screening rules rather than confirmed contamination. Using 33,004 label-applicable samples, we excluded Pb concentration, screening thresholds, label-source identifiers, and spatial split identifiers from the predictor set. Candidate models were selected by grouped out-of-fold evaluation in the development set and tested in held-out 50 km × 50 km spatial blocks. CatBoost achieved an outer-test PR-AUC of 0.4300. Under a 15% verification budget, the 1036-sample priority list captured 330 of 441 samples classified as high risk by the screening rules (74.83%), with a 31.85% within-list rate and 4.99-fold enrichment. Restricting development to the dominant USGS component retained 73.51% top-15% capture, with strong rank agreement to the integrated model (Spearman ρ = 0.8977). Sensitivity analyses separated stable, rule-sensitive, proxy-sensitive, and dual-sensitive candidates. The resulting ranking supports phased verification within the evaluated rule and data domain; broader transfer requires independent confirmation.