Conformal Prediction Sets for Artificial Intelligence Agriculture Crop Disease Triage Under Shifting Pathogen Priors: A Computer Coverage-Guarantee Evaluation

Conformal prediction has been proposed as a distribution-free method for equipping crop-disease classifiers with prediction sets that carry a finite-sample coverage guarantee. The guarantee, however, is marginal: it holds on average over the calibration and test distribution, and it can fail when the pathogen prior shifts between the two. A replication-based evaluation was conducted in which 600 data sets of N = 500 were drawn from a declared data-generating process with intercept 12.0, exposure 0.42, group -3.1, severity 1.35 and error SD 4.2, and the ordinary least squares estimator was characterised under the stated process. The estimated bias was 0.0658 for the intercept and -0.00103 for exposure, with root-mean-square errors of 0.87117 and 0.01522 respectively; empirical coverage of the nominal 95% confidence intervals was 1.0 for both parameters. Marginal coverage is an inadequate certificate for triage under prior shift, derives a prior-weighted coverage functional that makes the shift explicit, and shows that the coverage guarantee can be restored by likelihood-ratio reweighting of the conformity scores when the shift is known or estimable. The contribution is a formal statement of the conditions under which conformal triage sets remain valid for agricultural deployment, together with a diagnostic based on the empirical coverage of the reweighted sets.