Causal Mediation Decomposition for Artificial Intelligence Agriculture Fertilizer Recommendation Under Nutrient Runoff Confounding: A Computer Natural-Direct-Effect Estimation Evaluation

Fertilizer recommendation systems that learn from field data must separate the agronomic effect of a recommended rate from the effect that operates through nutrient runoff, because runoff both responds to application intensity and feeds back into yield through leaching and eutrophication pathways. Standard supervised recommenders estimate a total effect and therefore cannot distinguish a rate that raises yield directly from one that raises yield while exporting nutrients downstream. No observational data are used or claimed: the analysis characterises the design under a declared variance structure with between-cluster standard deviation 6.0, residual standard deviation 4.5 and intraclass correlation 0.64, across 2000 replications per design cell for cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points. The design attains 80 per cent power at 56 clusters for a 5 percentage point effect. The contribution is a mediation-aware evaluation protocol that makes the runoff channel an explicit estimand rather than an unmodelled confounder, together with the design parameters a field evaluation would require.