Benchmarking Data-Drift Detection Latency for Computer Agriculture Disease Classifiers Under Gradual Leaf-Symptom Onset: An Artificial Intelligence Alarm-Delay Bound for Fungicide-Timing Safety

Gradual leaf-symptom onset in crop disease creates a silent failure mode for computer-vision classifiers deployed in smallholder fungicide-timing decisions: the classifier's input distribution drifts while its decision boundary remains fixed, and the interval between drift onset and alarm constitutes an unmonitored exposure window. the present analysis formalises that interval as a detection-latency bound and characterises its operating characteristics under a before-after-control-impact design. No observational data are used or claimed: the analysis characterises the design under a declared variance structure with between-cluster SD = 6.0, residual SD = 4.5 and intraclass correlation = 0.64, evaluated 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 requires 54 clusters to detect a 5-percentage-point drift at 80% power. Estimated power curves, minimum detectable effect curves, coefficient intervals, kernel densities, cluster-power scatter and residual diagnostics are reported. The contribution is a design-level alarm-delay bound that converts a classifier's statistical detectability into a fungicide-timing safety margin, and a demonstration that the bound is governed by the intraclass correlation rather than by the nominal effect size alone.