Conformalized Quantile Inference for Computer Agriculture Irrigation Scheduling Under Sensor Calibration Drift: An Artificial Intelligence Coverage-Regret Evaluation

Soil-moisture sensors deployed for irrigation scheduling drift after field calibration, and the resulting bias propagates into every decision that treats a sensor reading as a state estimate. The scheduling rule is then defined as a regret-minimising threshold policy over the lower conformal bound, and a coverage-regret decomposition is derived that separates the regret attributable to interval miscoverage from the regret attributable to threshold suboptimality. A replication-based evaluation of the ['Intercept', 'exposure', 'group', 'severity'] estimator was conducted: 600 data sets of N = 500 were drawn from a declared data-generating process (intercept 12.0, exposure 0.42, group -3.1, severity 1.35, error SD 4.2), and the ordinary least squares estimator was characterised under that process. Estimated bias was -0.02082 for the intercept and 0.00049 for exposure; root-mean-squared error was 0.89005 and 0.01607 respectively; empirical coverage of the nominal 95% intervals was 1.0 for both parameters. The results indicate that the estimator is essentially unbiased for the exposure coefficient and that nominal coverage is attained, which supports the use of the conformal layer as a drift-robust wrapper rather than as a replacement for the underlying regression. Coverage-regret, not point accuracy, is the correct evaluation target for irrigation inference under calibration drift.