Post-Deployment OOD-Detection Evaluation for Computer Agriculture Pest-Classifiers Under Incremental Species Invasion: An Artificial Intelligence Open-Set Risk Bound

Pest classifiers deployed in computer agriculture are trained on a closed set of species, yet the field presents an open set in which new species arrive incrementally. 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 = 0.640. Power was estimated across 2000 design cells spanning cluster counts of 8, 16, 24, 32 and effect sizes of 3, 5, 8 percentage points. The design required 52 clusters to detect a 5 percentage-point effect at 80% power. The bound separates the OOD-detection problem into a closed-set calibration term and an open-set drift term, and shows that the drift term is governed by the rate of species arrival rather than by classifier capacity. The BACIP estimator is shown to be consistent for the drift term under the declared variance structure, and the power analysis identifies the cluster count at which the design becomes informative. The contribution is a formal risk bound that makes explicit the conditions under which a pest classifier's post-deployment OOD detection can be certified, together with a design-level evaluation of the statistical power needed to estimate the bound's components in the field.