Subgroup-Invariant Evaluation for Computer Agriculture Yield Forecasters Under Shifting Agro-Ecological Zones: An Artificial Intelligence Fairness-Audit Bound

Yield forecasters deployed in computer agriculture are typically evaluated on pooled error, yet their errors are known to concentrate in particular agro-ecological zones, and the zones themselves shift as climate and cropping patterns change. The bound is derived from a cluster-level before-after-control-impact design in which the estimand is the zone-specific change in forecast error, and it is shown to control the probability of falsely certifying subgroup invariance at a nominal level while retaining power to detect departures of practical size. A replication-based evaluation of the before-after-control-impact estimator was conducted 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 analysis indicates that 52 clusters are required to detect a 5 percentage-point departure at 80 per cent power, and that the minimum detectable effect at 80 per cent power falls steeply as cluster count rises from 8 to 32. The bound provides a defensible audit standard for subgroup-invariant evaluation and clarifies the cluster budgets under which such audits are informative rather than decorative.