Interpretability Bounds for Multimodal Tuberculosis Screening Classifiers Under Sputum Sample Transport Delay: A Computer Calibration Robustness Evaluation

Multimodal tuberculosis screening classifiers that fuse chest radiographs with sputum-based laboratory signals are calibrated on data collected under nominal transport conditions, yet sputum samples in peripheral settings routinely experience transport delays that shift the distribution of the laboratory modality. the present analysis formalises the interpretability bound as the maximum divergence between the classifier's attribution map under nominal and delayed transport that preserves the sign of the attribution for a specified fraction of the input space. Using a before-after-control-impact design with a declared variance structure (between-cluster SD = 6.0, residual SD = 4.5, intraclass correlation = 0.64), the operating characteristics of the design were characterised across 2000 replications per cell for cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points. The design achieved 80% power at 54 clusters for a 5-percentage-point effect. The interpretability bound is not a fixed property of the classifier but a function of the transport-delay distribution and the cluster-level variance decomposition, and that calibration robustness evaluations must report the bound alongside discrimination metrics to avoid overstating clinical readiness.