Calibration-Free Uncertainty Quantification for Artificial Intelligence Agriculture Crop Disease Triage Under Label Shift: A Computer Conformal Risk-Control Evaluation

Crop disease triage systems that route leaf images to agronomists must control the risk of missed infection when deployed across regions whose pathogen mix differs from the training distribution. Calibration-based uncertainty scores are widely used for this purpose, but they presuppose that the label marginal is stable, an assumption that fails under the label shift characteristic of seasonal and geographic deployment. The framework is evaluated through a replication-based design study of the before-after-control-impact estimator under a declared variance structure (between-cluster SD = 6.0, residual SD = 4.5, intraclass correlation = 0.64), with power estimated across 2000 design cells for cluster counts of 8, 16, 24 and 32 and effects of 3, 5 and 8 percentage points.54 clusters are required for 80% power at a 5-percentage-point effect, and that the conformal threshold attains finite-sample risk control without any calibration step. The contribution is a shift-robust triage rule whose guarantee is a coverage statement rather than a probability statement, together with a design-characterisation that tells field evaluators how many clusters their before-after-control-impact trial needs.