Artificial Intelligence Benchmarking of Neonatal Jaundice Screening from Smartphone Photographs Under Unmeasured Ambient Illuminant Spectra: A Healthcare Calibration-Robustness Evaluation

Smartphone-based neonatal jaundice screening promises low-cost triage where bilirubinometers are scarce, but its reliability depends on colour constancy under ambient illuminants that are never measured in deployment. the present analysis characterises the calibration-robustness of such screening as a design problem rather than a model problem. A before-after-control-impact (BACI) evaluation design is analysed under a declared variance structure with between-cluster SD = 6.0, residual SD = 4.5, and intraclass correlation = 0.64, using 2000 replications per design cell across 8, 16, 24, and 32 clusters and effect sizes of 3, 5, and 8 percentage points.54 clusters are required to detect a 5-percentage-point effect at 80% power, and that the minimum detectable effect at 80% power falls steeply as cluster count rises from 8 to 32. Kernel density and residual diagnostics indicate that the estimator is approximately unbiased but that power is dominated by between-cluster variance when illuminant spectra are unmeasured. The contribution is a formal statement of the calibration-robustness bound: under unmeasured illuminant spectra, the achievable power of any smartphone jaundice benchmark is capped by the ratio of between-cluster to residual variance, and this cap is independent of model architecture. The implication is that benchmark reporting for smartphone neonatal jaundice screening should report cluster-level design parameters alongside model metrics, because otherwise reported accuracy cannot be interpreted as screening performance.