Interpretability of Prototype-Based Artificial Intelligence Wound Classifiers Under Smartphone White Balance Drift: A Healthcare Calibration Fairness Evaluation

Prototype-based classifiers are attractive in wound care because their decisions can be inspected against learned exemplars, yet the interpretability they promise is conditional on the colour fidelity of the smartphone images that feed them. The problem is framed as a calibration-fairness question: when the same wound is photographed under different illuminants, does the classifier's nearest-prototype explanation remain stable, and does the stability loss fall disproportionately on darker skin tones? A formal account is developed in which white balance drift is modelled as a diagonal chromatic transformation acting on the input, and the induced displacement of prototype assignments is bounded in terms of the transformation's condition number and the local prototype margin. The analysis is methodological and theoretical: no wound images, patients or classifier outputs are reported. The contribution is a set of propositions that connect capture-side colour calibration to explanation-side fairness, together with a proposed audit protocol that treats white balance drift as a first-class source of interpretability failure rather than a preprocessing nuisance. The argument implies that prototype-based wound classifiers should be evaluated on explanation stability under illuminant shift, not only on classification accuracy, and that fairness audits should report per-skin-tone explanation agreement as a standard metric.