Distributionally Robust Optimisation for Artificial Intelligence Agriculture Input Subsidy Targeting Under Unobserved Plot Quality: A Computer Worst-Case Regret Evaluation

Subsidy targeting in smallholder agriculture must allocate inputs using covariates that are observed before planting, while plot quality that determines yield response remains unobserved. A nominal optimisation that treats the estimated yield-response function as true can therefore misallocate inputs whenever the deployment distribution of plot quality shifts away from the training distribution. The framework is formalised as a min-max problem over allocation weights, and a tractable convex reformulation is derived under a conditional moment model of yield response. Because no observational data are used or claimed, the evaluation is a design-based characterisation of operating characteristics under a declared variance structure: between-cluster standard deviation 6.0, residual standard deviation 4.5, intraclass correlation 0.64, and 2000 replications per design cell across cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points. The contribution is a computer-science formalisation that makes the price of robustness explicit and shows how the ambiguity radius governs the trade-off between nominal efficiency and worst-case regret in subsidy targeting.