MOSAIC-Age: Initial binding states and shared ageing trajectories establish a method-compatible predictive baseline...

Sequential-extraction and longitudinal studies document soil metal ageing, but differences in protocols, matrices and exposure histories limit cross-study assessment. MOSAIC-Age maps exact analytical protocols to method-compatible domains, equally weights independent experimental programs, and reconstructs shared reference trajectories anchored to the earliest measured state. Evidence spans 70 model-development and 23 registered holdout papers. P1 core training comprises six programs and 86 compositional states. Whole-program validation of 64 post-anchor states yielded RMSE 1.072 and 79.0% directional agreement. Two independent holdout programs yielded RMSE 0.654; ten external endpoints in a Cd/As research-family stratum yielded RMSE 0.275. In an agricultural-wastewater wetland experiment, predictions across 32 non-baseline states correlated with observations (r = 0.837), with RMSE 0.439 and 84.4% directional agreement. Treatment ordering was retained across eight 60 d endpoints (Spearman ρ = 0.738). Prediction intervals were assessed by coverage, width and proper scoring. Incorporating measured 60 d states improved 120 d forecasts over initial-anchor forecasts; persistence comparisons attributed improvement primarily to intermediate observations. The anchored baseline provides a shared temporal reference for state comparisons, while exposure and process observations identify local departures. MOSAIC-Age converts dispersed fractionation evidence into a computable, updatable baseline for relative mobility/activity screening and targeted monitoring, with uncertainty and applicability feedback.