Artificial Intelligence replication of Community Health Worker Diagnostic Image Triage Under Unmeasured Smartphone Camera White-Balance Drift: A Healthcare Fairness-Robustness Evaluation

Smartphone-based diagnostic image triage by community health workers is vulnerable to unmeasured white-balance drift, a camera-level nuisance that shifts colour distributions and can systematically distort the classifier outputs on which triage decisions depend. The framework is operationalised through a replication-based replication in which 600 data sets of N = 500 were drawn from a declared data-generating process with intercept 12.0, exposure 0.42, group -3.1, severity 1.35 and error SD 4.2, and the ordinary least squares estimator was characterised across replications. The replication returned estimated biases of 0.00145 for the intercept and 2e-05 for the exposure coefficient, root mean squared errors of 0.84328 and 0.0153 respectively, and empirical coverage of the nominal 95% confidence intervals of 1.0 for both parameters. These results establish a baseline against which drift-induced distortion can be quantified, and they show that under a correctly specified linear process the estimator is essentially unbiased and its intervals are conservative. Fairness-robustness evaluation of image triage must separate camera-level nuisance from label-level signal, and it proposes a decomposition that locates white-balance drift in the input distribution rather than in the decision rule.