Post-Hoc Isotonic Recalibration of Artificial Intelligence Agriculture Yield Forecasts Under District-Level Distribution Shift: A Computer Calibration-Error Evaluation

Machine-learning yield forecasts are increasingly used to allocate agricultural inputs and insurance, yet their probability statements are rarely calibrated to the districts where decisions are made. the present analysis treats post-hoc isotonic recalibration as a distribution-shift correction and evaluates its calibration-error properties under a before-after-control-impact design. No observational data are used or claimed: the analysis characterises the operating characteristics of the design under a declared variance structure with between-cluster SD = 6.0, residual SD = 4.5 and intraclass correlation = 0.640. Power was estimated across 2000 design cells. The evaluation shows that the before-after-control-impact interaction attains 80% power at 54 clusters for a 5 percentage-point effect, and that minimum detectable effects fall steeply as clusters increase from 8 to 32. Kernel-density and residual diagnostics indicate that calibration error is dominated by cluster-level variance rather than by residual noise, so recalibration gains are bounded by the number of districts contributing to the post-hoc fit. The contribution is a formal account of when isotonic recalibration can be expected to reduce calibration error under district-level shift, and a design rule linking cluster counts to detectable calibration improvements.