When the model becomes the measurement instrument

AI used upstream of analysis is part of the measurement process. Its false positives, false negatives, subgroup errors, version changes and context dependence can create misclassification, differential measurement error and artificial trends. It should therefore be validated as a measurement instrument for the epidemiological role it performs, not only benchmarked as a prediction model. Weavidence Insights. Independent scholarly publication; not externally peer reviewed. AI-assisted drafting and editorial development were used. Pietro Corona remains responsible for the final published claims, source selection, interpretation and approval.