Artificial Intelligence Benchmarking of Fundus Camera Diabetic Retinopathy Grading Under Unmeasured Mydriatic Drop Instillation Delay: A Healthcare Calibration-Transfer Evaluation

Diabetic retinopathy grading models are benchmarked on fundus photographs whose acquisition protocols are rarely recorded, and mydriatic drop instillation delay is a routine but unmeasured source of calibration drift. The design is characterised under a declared variance structure with an intraclass correlation of 0.64, between-cluster standard deviation of 6.0 and residual standard deviation of 4.5. Power was estimated across 2000 replications per design cell for cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points. Fifty-six clusters are required for 80% power at a 5 percentage point effect. The framework distinguishes delay-induced calibration shift from grading-rule error and specifies the conditions under which benchmark rankings are preserved or reversed. The contribution is a transfer-risk taxonomy and a design-theoretic account of when calibration-transfer evaluation can separate protocol artefacts from genuine model performance.