Counterfactual Yield Inference for Computer Agriculture Drought-Tolerant Seed Allocation Under Unobserved Plot Selection: An Artificial Intelligence Doubly-Robust Evaluation

Drought-tolerant seed allocation in computer agriculture is evaluated through counterfactual yield inference, yet unobserved plot selection confounds naive comparisons because farmers and algorithms assign varieties to plots on the basis of characteristics that are not recorded. The framework is formalised as a semiparametric estimator with a consistent variance estimator, and its finite-sample behaviour is examined through a theoretical analysis of the bias and variance trade-off. The analysis is used to illustrate how the doubly-robust estimator behaves when the outcome model is misspecified but the propensity model is correct, and vice versa. The contribution is a formal identification result, a finite-sample bias bound and a practical estimation algorithm for counterfactual yield inference in drought-tolerant seed allocation under unobserved plot selection.