Interpretability Audits of Computer Vision Fundus Graders Under Tropical Heat-Induced Lens Haze: A Healthcare Calibration Fairness Evaluation

Fundus cameras deployed in tropical clinics accumulate heat-induced lens haze that degrades image contrast in ways that differ across camera bodies, operators and sessions, yet interpretability audits of diabetic retinopathy graders typically assume a stationary image-formation process. The framework separates a grader's discriminative ranking from its probability calibration and defines a haze-conditional calibration gap as the divergence between calibration curves estimated under clear and hazed acquisition. A before-after-control-impact design is characterised analytically 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. The design requires 52 clusters to detect a 5-percentage-point calibration gap at 80% power. The contribution is a formal statement of when interpretability audits can distinguish haze-induced miscalibration from sampling noise, together with the audit quantities that must be reported for that distinction to be possible.