Cross-region weather models gave catch calibration slopes of 0.47 to 0.81 in four British moth samples in 1993-2015

We tested a weather model across English and Scottish moth sites and two periods from 1993 to 2015. Training used only the other region and the other period. Four pooled tests used 29,041 trap nights. Calibration slopes were 0.47 to 0.81, where one retains the trained model and zero gives uniform allocation among recorded nights. All four year-cluster intervals were below one conditional on trained coefficients. Only early England had a clear log-score gain over uniform allocation after four-test correction. A year bootstrap that refits training makes the early-English calibration interval include one. Site omission showed strong Scottish variation. Missing zero reports, selected sites, few years and training uncertainty limit inference. This diagnostic concerns recorded catch shares. It does not identify detection, population trends or causal climate effects. No pollution was tested. Code, aggregate evidence, source hashes and a verified reproduction route are included.