Artificial Intelligence Adversarial Robustness Certification of Community Clinic Malaria Rapid Diagnostic Reading Under Unmeasured Lighting Degradation: A Healthcare Safety Guarantee Evaluation

Rapid diagnostic tests for malaria are read by community health workers under lighting conditions that are rarely measured, and the resulting uncertainty is not represented in the certification claims made for the artificial-intelligence readers that increasingly assist them. the present analysis treats that gap as a certification problem rather than a data problem. It develops a distributionally robust certification framework in which the adversarial perturbation set is defined over a lighting-degradation operator acting on the image, and it derives a safety guarantee that holds uniformly over that set without requiring the deployment lighting distribution to be known. The framework is evaluated through a replication-based power analysis of the before-after-control-impact estimator that would be used in a field certification trial, with a declared variance structure (between-cluster SD = 6.0, residual SD = 4.5, intraclass correlation = 0.64) and 2000 replications per design cell across cluster counts of 8, 16, 24 and 32 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 intraclass correlation of 0.64 dominates the design burden. The contribution is a certification logic that separates the lighting-robustness guarantee from the trial design that would audit it, so that neither is allowed to substitute for the other.